Orthogonal Factors Describing Primary and Spatial Sensations of the Sound Field in a Concert Hall

Yoichi Ando

Graduate School of Science and Technology, Kobe University Rokkodai, Nada, Kobe 657-8501 Japan

Subjective preference of the sound field in a concert hall is described based on the model of human auditory-brain system. The model consists of the autocorrelation function (ACF) mechanism and the interaural crosscorrelation function (IACF) mechanism for signals arriving at two ear entrances, and the specialization of human cerebral hemispheres [Ando, Architectural Acoustics, AIP/Springer, 1998]. From this view point, primary sensations such as pitch or missing fundamental, loudness, timbre, and in addition duration sensation which is introduce here as a fourth are well described by the temporal factors extracted from the ACF associated with left hemisphere. And, spatial sensations such as apparent source width (ASW) and subjective diffuseness are described by the spatial factors extracted from the IACF associated with the right hemisphere

ORTHOGONAL FACTORS peak. Usually, there are certain correlation between τ τ φ φ n and n +1, and between n and n+1; Primary sensations and spatial sensations as well as (3) Effective duration of the envelope of the subjective preference for sound fields are well normalized ACF, τe, which is defined by the ten- described by a model of the auditory-brain system. The percentile delay and which represents a repetitive model includes autocorrelation function (ACF) and feature or reverberation containing the sound interaural crosscorrelation function (IACF) source it. mechanisms [1,2]. Important evidences supporting this (a) 1 model were discovered in relation to the auditory-brain activity [2]. This article reviews that primary φ sensations and spatial sensations are mainly described 1 by temporal and spatial factors extracted from the ACF ) τ

( 0 and the IACF, respectively. p φ

τ Factors extracted from the ACF 1

-1 The ACF is defined by 0 Delay time τ [ms] +T Φ τ 1 τ (1) (b) 0 τ p( ) = p'(t)p'(t+ )dt e 2T -T

[dB] -5 where p’(t) = p(t)*s(t), s(t) being the ear sensitivity, ) | which is essentially formed by the transfer function of τ ( p physical system to oval of cochlea. For convenience, φ -10 s(t) may be chosen as the impulse response of an A- weighted network [1,2]. The ACF and the power log | density spectrum contain the same information. There -15 are four factors, which can be extracted from the ACF: 020100 0 Delay time τ [ms] (1) Energy represented at the origin of the delay, Φ (0); p FIGURE 1. Definition of independent factors other than (2) Fine structure, including peaks and delays (Figure Φ(0) extracted from the normalized ACF. (a) Values of 1a). For instance, τ and φ are the delay time and τ φ 1 1 1 and 1 for the first peak; (b) The effective duration of τ φ τ the amplitude of the first peak of ACF, n and n the ACF e is obtained practically by the extrapolation being the delay time and the amplitude of the n-th of the envelope of the normalized ACF during the decay, 5 dB initial (b).

The normalized ACF is defined by

φp(τ) = Φp(τ) /Φp(0) (2)

As a manner shown in Figure 1b, the value of τe is obtained by fitting a straight line for extrapolation of delay time at –10 dB, if the initial envelope of ACF decays exponentially. Therefore, four orthogonal and temporal factors that can be extracted from the ACF are Φp(0), τ1, φ1, and τe .

Auditory-Temporal Window

In analysis of the running ACF, of particular interest FIGURE 2. Definition of independent factors IACC, is so called an “auditory-temporal window”, 2T in τIACC and WIACC extracted from the normalized IACF. Equation (1), that must be determined. Since the initial τ part of ACF within the effective duration e of the ACF contains the most important information of the signal, PRIMARY SENSATIONS thus the recommended signal duration (2T)r is given by Loudness

(2T)r ≈ K1(τe)min [s] (3) Let us now consider primary sensations. Loudness sL where (τ ) is the minimum value of τ obtained by is given by e min e analyzing the running ACF, K1 being the constant sL = f[Φp(0), τ1, φ1, τe, D] (6) around 30 [7]. The running step (Rs) is selected as K2(2T)r, K2 being selected, say, in the range of 1/4 – 3/4. where D is the duration of sound signal as is represented by musical notes. It is worth noticing that τ Factors extracted from the IACF the value of 1 corresponds to pitch of sound and/or the missing fundamental as discussed below. Since the The IACF is given by sampling frequency of the sound wave is more than the twice of the maximum audio frequency, the value

+T 10logΦ (0)/Φ (0)ref is far more accurate than the Leq 1 Φlr(τ) = p'l(t)p'r(t+τ)dt (4) which is measured by the sound level meter. 2T -T Scale values of loudness within the critical band were obtained in paired-comparison tests (with filters where p’ (t) = p(t) *s(t), p(t) being the sound with the slope of 1080 or 2068 dB/octave) under the l ,r l,r l,r Φ pressure at the left- and right-ear entrances. The condition of a constant p(0) [2,4]. Obviously, when normalized IACF is given by sound signal has the similar repetitive feature, τe becomes a great value, as like a pure tone, then the 1/2 φlr(τ) = Φlr(τ)/[Φll(0)Φrr(0)] (5) greater loudness results. Thus a plot of loudness versus bandwidth is not flat in the critical band. This where Φll(0) and Φrr(0) are autocorrelation functions contradicts previous results of the frequency range (τ = 0) or sound energies arriving at the left- and right- centered on 1 kHz [5]. ear entrance, respectively. Spatial factors extracted from the IACF are defined in Figure 2 [2]. Pitch In analyzing the running IACF, 2T is selected by Equation (3) also. For the purpose of spatial design for The second primary sensation applying the ACF is sound fields, however, longer values of (2T)r may be the pitch or the missing fundamental of the noise. It is useful, because it is essentially time independent. given by

sP = f[Φp(0), τ1, φ1, τe , D] (7)

When a sound signal contains only a number of Table 1. Primary sensations in relation to factors extracted harmonics without the fundamental frequency, we hear from the autocorrelation function and the interaural crosscorrelation function. the fundamental as a pitch. This phenomenon is well explained by the delay time of the first peak in the

ACF fine structure, τ1 [6,7]. According to experimental Factors Primitive Sensations results on the pitch perceived when listening to a) bandpass noises without any fundamental frequency, Loudness Pitch Timbre Duration the pitch sp is expressed by equation (7) as well, under the condition of a constant Φπ(0). The strength of the ACF Φp(o) X x X X pitch sensation is described by the magnitude of the τ φ 1 X X X X first peak of the ACF, 1. For a signal of short duration, φ x X X X factor D must be taken into account. 1 τe X x X x D xb) xb) Xb) X Timbre

The third primary sensation, timbre that includes X and x : Major and minor factors influencing the corresponding response, respectively. pitch, loudness, and duration, might be expressed by a). Timbre in relation to all of temporal and spatial factors is under investigation. b). It is suggested that loudness, pitch and timbre should be sT = f[Φp(0),τe, τ1, φ1, D] (8) examined in relation to the signal duration. It is worth noticing that the intelligibility of single τ syllables as a function of the delay time of single s = f[LL, IACC, IACC, WIACC] (10) reflection is well be calculated by the four orthogonal factors extracted from the running ACF analyzed for where the piece between consonant and vowel sounds [7]. A Φ Φ recent investigation, clearly show that timbre or LL = 10 log [ p(0)/ (0)ref] (11) dissimilarity judgment is an overall subjective response Φ Φ Φ 1/2 Φ Φ similar for the subjective preference of sound fields in And p(0) = [ ll(0) rr(0)] , and ll(0) and rr(0) concert hall. being ACFs at τ = 0 (sound energies), of the signals arriving at the left and right ear-entrances. In four orthogonal factors in Equation (10), the interaural Duration delay time, τ , is a significant factor in determining IACC the perceived horizontal direction of the source. A The forth-primitive sensation, which is introduced well-defined direction is perceived when the here, is the perception of signal duration, which is normalized interaural crosscorrelation function has one given by [12,13] sharp maximum, a high value of the IACC and a

narrow value of the W , due to high frequency s = f[Φ (0), τ , φ , τ , D] (9) IACC D p 1 1 e components. On the other hand, subjective diffuseness

or no spatial directional impression corresponds to a One of experimental results has been expressed in τ φ low value of IACC (< 0.15) [9]. relation to 1, 1, and D [8]. Table 1 indicates Of particular interest is that, for the perception of a summarization of primary sensations in relation to sound source located in the median plane, the temporal factors extracted from the ACF and physical signal factors extracted from the ACF of sound signal duration D. arriving at the ear-entrances may act as cues . It has

been shown that three factors, τe, τ1, and φ1 as a SPATIAL SENSATIONS function of the incident angle greatly differ, but few differences may be found in the head-related transfer Directional Sensation functions [10]. A remarkable finding is that there are neural activities at the inferior colliculus corresponding to the If Φll(0) ≈ Φrr(0), then the perceived direction of a noise source in the horizontal plane is assumed to be IACC and sound energies for sound signals that described as arriving at the two-ear entrance [11]. Also, it is discovered that the LL and the IACC are dominantly associated with the right cerebral hemisphere, and the

∆ temporal factors, t1 and Tsub, the sound field in a room Table 2. function (IACF). Spatial sensations in relation to are associated with the left [2]. factors extracted from the autocorrelation function (ACF) and the interaural crosscorrelation Subjective Diffuseness Factors Spatial Sensations The scale value of subjective diffuseness is assumed ASW Subjective Image Horizontal Vertical to be given by Equation (10). In order to obtain the Diffuseness Shift Direction Direction scale value of subjective diffuseness, paired- comparison tests with bandpass Gaussian noise, varying the horizontal angle of two symmetric ACF τ reflections have been conducted. Listeners judged 1 X φ X which of two sound fields were perceived as more 1 τe X diffuse, under the constant conditions of LL, τIACC, and IACF - - X X x WIACC [12]. The strong negative correlation between Φll(0) - - X X x the scale value and the IACC can be found in the Φrr(0) results with frequency bands between 250 Hz - 4 kHz. LL X X - - - τ The scale value of subjective diffuseness may be well IACC x x X X x WIACC X X X X x formulated in terms of the 3/2 power of the IACC in a IACC X X X X x manner similar to the subjective preference for the sound field, i.e., X: Major factors influencing the corresponding response. β 1/2 LL = 10 log [Φ(0)/Φ(0)fref], where Φ(0) = [Φll(0) Φrr(0)] ; Sdiffuseness ≈ - α(IACC) (12) ASW: Apparent source width.

α ≈ β ≈ where coefficients 2.9 and 3/2. REFERENCES Apparent Source Width (ASW)

1. Y. Ando 1985 Concert hall acoustics, Springer-Verlag, It is considered that the scale value of apparent Heidelberg. source width (ASW) is given by equation (10) as well. 2. Y. Ando 1998 Architectural acoustics, blending sound For a sound field with a predominately low frequency sources, sound fields, and listeners. AIP Press/Springer- range, the long-term IACF has no sharp peaks for the Verlag, New York. delay range of | τ | < 1 ms, and WIACC becomes wider. 3. K. Mouri, K. Akiyama and Y. Ando, J. Sound Vib., 241, Clearly, the ASW may be well described by factors, 87-95 (2001). 4. S. Sato, T. Kitamura, H. Sakai and Y. Ando, J. Sound IACC and WIACC [7], under the conditions of a constant LL and τ = 0. The scale values of ASW were Vib., 241, 97-103 (2001). IACC 5. E. Zwicker, G. Flottorp, and S.S. Stevens, J. Acoust. Soc. obtained by paired-comparison tests with ten subjects. Am., 29, 548-557 (1957). The listening level affects ASW [13], therefore, the 6. M. Inoue, Y. Ando and T. Taguti, J. Sound Vib., 241, total sound pressure levels at the ear canal entrances of 105-116 (2001). sound fields were kept constant at a peak of 75 dBA. 7. Y. Ando, H. Sakai and S. Sato, J. Sound Vib., 232, 101- Listeners judged which of two sound sources they 127 (2000). perceived to be wider. The results of the analysis of 8. K. Saifuddin, H. Sakai, and Y. Ando, J. Sound Vib., 241, 117-127 (2001). variance for the scale values sASW indicates that both of 9. P. Damaske and Y, Ando, Acustica, 27, 232-238 (1972). factors IACC and WIACC are significant (p < 0.01), and contribute to the s independently, thus 10. S. Sato, V. Mellert and Y. Ando, Sound Vib., 241, 53- ASW 56 (2001). 11. Y. Ando, K. Yamamoto, H. Nagamastu and S.H. Kang, ≈ 3/2 1/2 sASW a(IACC) + b(WIACC) (13) Acoust. Letters, 15, 57-64 (1991). 12. Y. Ando and Y. Kurihara, J. Acoust, Soc. Am., 80, 833- where coefficients a ≈ -1.64 and b ≈ 2.44. Table 2 836 (1982). indicates a list of spatial sensations with their 13. M.V. Keet, Proc. 6th Intern. Congr. Acoust., Tokyo, significant factors extracted from the IACF. Paper E-2-4 (1968). Fundamental subjective responses for the sound field in a concert hall may be described by all of significant orthogonal factors. For example, the scale value of subjective preference is well described by four orthogonal factors, i.e., LL, IACC, ∆t1 and Tsub [1,2]. The Preferred Acoustic Parameters for a Javanese Gamelan Performance Hall

J. Sarwonoa,b and Y.W. Lama

aSchool of Acoustics and Electronic Engineering, University of Salford, Brindley Building, Meadow Road Site, Salford M7 9NU, UK. E-mail: [email protected] bEngineering Physics Department, ITB, Jl. Ganesa 10 Bandung 40132, Indonesia.

This paper discusses the application of a method based on human subjective preference to the acoustic design of a Javanese gamelan performance hall. Some important distinctions between Javanese gamelan ensembles and Western classical orchestra are the tuning system, orchestral blending process, and technique of playing. The results of subjective preference test using the rank order method showed that the subjects preferred 30 ms for ITDG, 600 ms for RT, and the smallest value of IACC. These results, except for the IACC, agree with the acoustic parameters from the room responses measured in a traditional pendopo in Indonesia, which is not a common concert hall but an open-sided hall.

INTRODUCTION echoic chamber. A computer-based analysis has been used to obtain the most appropriate gendhing for the Javanese gamelan is one of the Indonesian whole subjective preference test, while computer traditional music ensembles. There are several simulation process was mainly used for preparing the important differences between the gamelan and the test samples. In situ measurements were conducted in a Western symphony orchestra including tuning systems, pendopo in Indonesia to provide comparison for the orchestral blending systems, and playing technique. subjective preference tests. According to Ando[1], by using human preference All the subjective preference tests were carried out approach through a psychoacoustic test, four in an anechoic chamber, using a configuration of 7 orthogonal factors for designing concert hall can be loudspeakers to simulate several sound field conditions determined. Those four factors are the listening level to be judged by listeners. All the listeners were (LL), the initial time delay gap (ITDG), the subsequent university students with several nationalities, inclusive reverberation time (RT), and the Inter-Aural Cross- all genders. The subjective preference test has been Correlation (IACC). So far, this theory has mostly been carried out using the rank order method. applied for designing concert halls for Western A studio recording gendhing from the closing part of τ classical music. Kebogiro Glendeng, with minimum e = 27.59 ms (2T This paper will discuss an application of the = 2 s, interval 100 ms), was used in the subjective approach to design the preferred acoustic conditions preference test. The duration of the stimulus was 9.3 s. for performing Javanese gamelan in an enclosed hall. All stimuli were stored in a PC, which was also Three preferred parameters, ITDG, RT and IACC will functioned as stimuli player. Seven identical be discussed in this paper. Measurement data from a loudspeakers were used to produce the sound. All pendopo, an open-sided hall where Javanese gamelan loudspeakers were placed at distance of 1.35 m from usually played, in Indonesia will be provided as the listener. The horizontal angles of the loudspeakers comparison. were 0o, ±45o, ±67.5o, and ±135o. The vertical angles of the loudspeakers were 0o, except the rear METHOD AND EXPERIMENT SETUP loudspeakers for the ITDG and RT tests, which were elevated 6o, relative to the subject's ears. The detail The research combines three major methods, a configuration is shown in Table 1. computer based analysis and simulation, in situ measurements and subjective preference test in an-

Table 1. Detail of Test Configuration Direct Refl. Reverb. Refl. Reverb. Stimuli Listening Subject Test Sound Amplitude Amplitude Level o o o o ITDG 0 ±45 ±67.5 , ±135 1 dB -3 dB 15, 30, 50, 80, 160 ms 73 dBA 6 RT 0o ±45o ±67.5o, ±135o -1 dB 2 dB 0, 0.45, 0.6, 1.2, 2.5, 4.5 s 73 dBA 17 o o o o IACC 0 ±45 ±67.5 , ±135 vary vary 0.3, 0.4, 0.5, 0.75, 1 73 dBA 10 RESULTS AND DISCUSSION 1000 35

900 It was shown that there was a low value preference 30 for ITDG (Figure 1) as well as for RT (Figure 2), with 800 the most preferred value of 30 ms and 600 ms, 700 25 600 respectively. This means that the subjects preferred 20 500 good clarity with an intimate sound field for listening 15 400 RT (ms) to Javanese gamelan in an enclosed hall. These results ITDG (s) 300 agree with the ITDG and RT of pendopo Puro 10 200 Mangkunegaran[2], as shown in Figure 3. It shows the 5 100

0 0 centre 10 11 15 king 5 Measurement points

4 RT ITDG

3 FIGURE 3. ITDG and RT of Pendopo Mangkunegaran

2 Rank Order 5 1

4 0 15 30 50 80 160 ITDG (ms) 3

FIGURE 1. Preference for ITDG 2 Rank Order

6 1

5 0 0.3 0.4 0.5 0.75 1

4 IACC

FIGURE 4. Preference for IACC 3 Rank Order 2 CONCLUSION

1 0 450 600 1200 2500 4500 The preferred parameters for Javanese gamelan performance hall were 30 ms for ITDG, 600 ms for RT (ms) RT, and the smallest value of IACC. These agree with FIGURE 2. Preference for RT the acoustic parameters, except for the IACC, from the room responses measured in a traditional pendopo in Indonesia, which is not a common concert hall but an ITDG and RT of the pendopo at 5 measurement points, open-sided hall. including the centre of the hall (centre), the audience area (10, 11, 15), and the VIP area (king). Figure 4 shows that the lower the IACC the higher REFERENCES the subjective preference. This shows that a 1. Ando Y, "Architectural Acoustics", Springer Verlag, spaciousness and enveloping sound field is preferred New York, 1998. for listening to Javanese gamelan in an enclosed hall. 2. Sarwono, J. and Lam, Y.W., "The Acoustics of a However, this is not in agreement with the measured Pendopo: A Typical Open-sided Hall for Javanese IACC of pendopo Puro Mangkunegaran, (IACC = 1) Gamelan Music Performance", in proceeding of IoA, as it is an open-sided hall. 2000, Volume 22 Pt 2, pp. 305 - 313. The Application of Neural Network Analysis to Auditorium Acoustics

F. Fricke

Department of Architectural and Design Science, University of Sydney, NSW 2006, Australia. [email protected]

Neural network analysis (NNA) is a relatively new research and design tool that has been used in many fields from structural engineering to finance. So far very little use of the technique has been made in architectural acoustics. In this paper the NNA technique is outlined and examples of its use in auditorium acoustics are given to demonstrate its potential. These include the prediction of reverberation time and sound levels in auditoria and the acoustic quality of halls using both acoustic and physical parameters as inputs. The advantages and limitations of neural network analysis are also outlined.

INTRODUCTION application to a number of architectural acoustics issues has been described in a several papers by There are at least two approaches to the study of Fricke eg [5],[6] and Nannariello eg [7],[8]. concert hall and auditorium acoustics. One is academic The method is based on the way the brain works and the other design oriented. The academic approach is where neurons are connected by synapses. In a directed at finding out what it is that makes concert halls simple NNA the inputs (eg length and height of a good and what influences opinions about the acoustics of room) neurons are interconnected to a layer of halls. It is also about measuring and calculating various “hidden” neurons which in turn are connected to an acoustic quantities in halls and trying to apply results of output (eg the reverberation time or “acoustic perception experiments, carried out in anechoic rooms, quality” of a room) neuron. The network is trained, to more complex situations such as that which exist in using data (cases) from existing situation where the concert halls. In the second approach the architect or inputs and outputs are known. The error between the designer wishes to define the acoustics of a space in actual and predicted values of the output is terms of its size, shape and surface finishes. minimised by systematically changing the weights on While the approaches of Beranek [1], Ando [2] and the connections between the neurons. others shows great understanding of the academic The advantages over other approaches are that requirements these approaches do not give designers the NNA can handle more than 6 input variables (usually tools they want. These tools are simple rules of thumb considered the maximum possible number for a that ensure excellent acoustics. Such simple rules almost conventional analytical approach) and can deal with certainly do not exist but more complex ones possibly non-linear relationships. Its disadvantages are that it do. For instance, the most basic rule of thumb used is never possible to determine whether an optimal seems to be the volume per seat even though the volume solution has been found and when a solution has been per seat varies between good halls (Boston Symphony found it cannot easily be used in the form of an Hall has a V/N of 7.14 while Meyerson Hall has 11.6.). equation though it can be easily used in a spreadsheet A more complex rule may, for example, involve the format. Often there are not enough cases available to optimum volume per seat as a function of the length of accurately train, verify and test a network and the the hall. Ultimately the aim of the present work is to validity of the analysis is only within the range of the investigate whether such complex rules exist and if so, to input variables. Also, where there are more than 6 present them in a designer-friendly form. inputs, it is very difficult to represent the output graphically or to produce rules of thumb from the analysis. NEURAL NETWORK ANALYSIS NNA OF THE ACOUSTIC QUALITY Very briefly, neural network analysis (NNA) is a computer-based technique which learns to recognize OF ROOMS patterns. These patterns are usually in numerical data but could be in the juxtaposition of pixels or the pitch of Of the two approaches tried for the prediction of notes. The general technique and its applications have acoustic quality of rooms the “acoustic input” been described in many texts eg [3],[4] and its approach [6] gives better results (Standard Deviation Ratio, SDR ≈ 0.2) than the “geometrical input” approach are to be predicted falls within range of the training [5] (SDR ≈ 0.9). This is not surprising given the large data for the neural network. number of geometrical inputs required to define an There are limitations on the method and if NNA is auditorium (though many of them are related to one to be a success there is a need for a data base on the another). The geometrical approach required 10 inputs web where information can be made available to (V, S, N, L, W, H, SDI, MRA, SH and SE) while the everyone. This is necessary as it is doubtful if any acoustic approach required only 6 (5 of Beranek’s input one person is ever going to be able to undertake all parameters – EDT, G, IACC, TI, BR and SDI - and the measurements needed on halls in order to carry either N or V) where V = room volume, S = room out satisfactory neural network analyses. surface area, N = number of seats, L, W and H are the As a final comment it must be stated that NNA maximum length, width and height respectively, SDI = should not be considered as a new branch of surface diffusivity index, MRA = mean rake angle of architectural acoustics but rather as a new fertiliser seating, SH is stage height and SE = stage enclosure. which may help the existing branches bear more One modified geometrical approach which has given fruit. useful results involves categorising halls into two groups; those with an AQI of 0.7 or greater and those REFERENCES with an AQI of less than 0.7. With this approach there is > 90% success rate using N, L/W, H/V1/3, MRA and SDI/SE as inputs. 1. Beranek, L. L., Concert Halls and Opera Another approach is based on Nannariello’s work [8] Houses, Acoustical Society of America, Woodbury, in which acoustical parameters, such as IACC and RT, NY, 1996 are obtained from geometrical inputs. These can then be 2. Ando, Y., Concert Hall Acoustics, Springer- used to calculate AQI. The efficacy of this method Verlag, Berlin, 1985. should not be in doubt given Nannariello’s results for G, 3 Fausett, L., Fundamentals of Neural Networks: RT, and IACC, (and the certainty that room acoustic Architecture, Algorithms & Applications, Prentice parameters are dependent on size, shape and surface Hall, New Jersey, USA, 1994. finishes of rooms), but the final analysis has yet to be 3. Statistica Neural Networks, (1999) Technical carried out. Manual Version 4, StatSoft Inc., Tulsa, OK. 4. Fricke, F. R. & Han, Y. H., (1999), A Neural Network Analysis of Concert Hall Acoustics, DISCUSSION AND CONCLUSIONS Acustica, 85, 113- 120. 5. Fricke, F. R., (2000), Concert Hall Acoustic NNA can be used to predict the acoustic quality of a Design: An Alternative Approach, Building concert hall or an auditorium though the accuracy of the Acoustics, 7, 233-246. “geometrical” approach leaves something to be desired. 6. Nannariello, J. & Fricke, F. R., (1999), The Both the “geometrical” and “acoustical” NNA Prediction of Reverberation Time Using Neural approaches are useful in understanding the influences on Network Analysis, Applied Acoustics, 58 (3), 305- the acoustic quality of auditoria and giving an estimate 325. of acoustic quality early in the design process. It appears 7. Nannariello, J. & Fricke, F. R., (2001), likely that much better predictions of acoustic quality, Introduction to neural network analysis and its using geometrical inputs and more complex networks, application to building services engineering, will be developed soon. Once such a network has been Building Services Engineering Research & developed and the network embedded in a spreadsheet Technology Journal, 22, 61-71 for designers to use. 8 Nannariello, J. & Fricke, F. R., (2001), The Likewise, NNA can be used to predict acoustical Prediction of Reverberation Time Using suitable quantities in auditoria such as RT (or EDT), IACC, G, Neural Networks, Proceedings 17 ICA, Rome BR and TI provided that the space in which the quantities Objective evaluations of chamber music halls in Europe and Japan. Takayuki Hidaka*, Noriko Nishihara*

* Takenaka R&D Institute, 1-5-1, Otsuka, Inzai, Chiba 270-1395, Japan

Abstract: The room acoustical parameters - Reverberation time RT, early decay time EDT, clarity C80, strength G, initial time delay gap ITDG, and interaural cross-correlation coefficient IACCE, were measured in 18 major chamber music halls in Europe and Japan employing the procedure in accordance with ISO 3382 [1]. By combining architectural data, the intrinsic parameters for the acoustics of chamber music halls are examined. INTRODUCTION volumes of the latter are about 40% larger. The reason for the size differences appears to come from the fact For symphony halls and opera houses, the results of that modern architects prefer medium-upholstered measurements of current room acoustical parameters chairs for greater comfort. Because such chairs have been reported in the literature [2,3]. There are absorb more sound, even when occupied, the room only limited numbers of similar studies on chamber volume is larger in modern halls so as to adjust the RT music halls [4]. There is no assurance whether to the volumes shown. The approximate equation = ⋅ existing data or design guidelines for large symphony with the form, RTM ,occ K V / S A , is plotted in Fig. halls are also suitable for smaller sized spaces, 1, where relevant K value falls between 0.13 and 0.14 therefore it seems meaningful to assemble the for chamber halls, similar to the value of 0.14 for acoustical data and to survey their features. In this symphony halls [2], and RT’s seem to converge to ca. paper, 9 highly-reputed halls of traditional design in 1.8 s. Europe and 9 major halls of contemporary design in C80, EDT : C80 and EDT are variables not independent Japan are compared and studied. from RT, but all these are very highly correlated. However, the subjective impression of clarity in MEASUREMENT RESULT AND SOME chamber halls is frequently of major concern. As DISCUSSION shown in Fig. 2, C80’s (occupied) may be classified into two groups, (3.5±0.4) and (0.1±1.6) dB. The The measured halls, which are regularly used for latter coincides with the optimal range for Mozart chamber music in each city, are listed in Table 1. music which was proposed by Reichardt et al. [7]. European and Japanese halls respectively can be Obviously every hall exceeds the lower limit of -1.5 classified as those of traditional style and those of dB. modern construction and materials. The seating G : Strengths G in dB for traditional and modern halls numbers, N, in these 18 halls vary from 207 to 844, are moderately different from each other, except for while the volumes, V, and reverberation times hall SG, with the largest capacity N=844 (Fig. 3). G (occupied) vary from 1070 to 8475 m3 and 0.9 to 2.0 s, L and GM of the former are respectively about 4.5 and 3 respectively. Many of them (15 out of 18 halls) are dB larger than the latter on average, which is probably shoebox, or at least have rectangle floor plans. The caused by the difference in volume, e.g., Beranek has suffix “L”, “M” and “3” associated with the measured shown G is proportional to 10log(EDT/V) [2]. quantities mean the average over 125/250 Hz, 500/1k BR : The bass ratios for occupied condition are Hz, and 500/1k/2k Hz, respectively. The occupied distributed from 0.87 to 1.12 and from 1.07 to 1.24 for values were transformed from measured unoccupied modern and traditional halls (median values are 1.02 values using the method in [5]. and 1.14), respectively, which are narrower ranges than The measurements were executed without audiences that of the concert halls, 0.92 to 1.45 in spite of the and with no instruments on the stage (sometimes a wider range of V/N. BR highly correlates with G piano existed at the corner of the stage). The L (r=0.8), although Bradley and Soulodre find that GL is measuring procedure is exactly the same as in [3,5] and more significant [8]. coincides with that of ISO 3382 [1]. The correlation [1-IACCE,80] : [1-IACCE] is also an independent matrix for the objective measures shown in Table 2 variable for chamber halls but the variation range is indicate that the independent parameters are RTM, G, extremely narrow, 0.67 to 0.77. This range is same as IACCE3, BR, and ITDG. This same correlation matrix the subjective difference limen by [9], namely it can be is also found in symphony halls and opera houses [3,6]. said that every chamber hall has similar binaural RT : The volume per person on average is 6.4 m3 for quality, provided [1-IACCE,80] is still valid for chamber traditional halls and 9.1 m3 for modern halls, thus the hall. This situation is quite different from that for a large symphony hall or opera house, where the varied within narrow range so that the integration limit variation range is 0.39 to 0.72. Physically, there are of IACCE should be reduced to the first 30 msec to very many lateral reflections within the first 80 msec in separate each hall suitably. Further research is every hall. If we assume that binaural correlation required to verify its subjective foundation. plays a significant role for sound quality in small halls, it is possible to separate them using [1-IACCE,30], REFERENCES where only the information within the first 30 msec is used (Fig. 4). Although there is no precise evidence [1] ISO 3382: 1997, “Acoustics - Measurement of the at the present moment as to why the integration should reverberation time of rooms with reference to other acoustical parameters.” [2] Beranek, L. L., Concert and be limited to 30 sec, the possibilities may be (1) early Opera Halls, Acoust. Soc. Amer, 1996. [3] Hidaka, T. and reflections after 20 msec may deteriorate the Beranek, L. L., J.A.S.A., 107, 368-383 (2000). [4] Barron, localization of stereophonic sound [10], and (2) M., Auditorium Acoustics and Architectural Design, E and FN audiences may relish more detailed information for Spon, London, 1993. [5] Hidaka, T., et al., J.A.S.A. 109, delicate chamber music. 1028-1042 (2001) [6] Hidaka, T., et al., J.A.S.A., 107, 340- 354 (2000). [7] Reichardt, W., et al., Acustica 32, 126- CONCLUSION 137 (1975). [8] Bradley, J. and Soulodre, G., J.A.S.A., 98, 2590-2597 (1995). [9] Cox, T. J., et al., Acustica 79, 27-41 (1993). [10] Bech, S., 100th Convention AES, Copenhagen RT, G, BR, ITDG, and IACCE are independent (1996). parameters in the chamber halls studied. However, 2.5 their values for traditional and for contemporary halls RTocc, M vs. V/SA have different ranges except for IACC . IACC Europe E,80 E,80 2.0 Japan VS

KM Table 1 Chamber music halls for which objective measurements are available. TI PM VB SGTM TD TH TC

VNV/SA RT occ, M EDTunocc, M BRocc. C803B GL GM 1-IACCE3 ITDG (sec) ZTVM m3 m sec sec - dB dB dB (80msec) msec 1.5 occ, M AC Amsterdam, Kleinersaal in Concertgebouw 2,190 478 9.4 1.25 1.49 1.21 1.5 13.7 12.9 0.69 17 TT RT AC BS Berlin, Kleinersaal in Schauspielhaus 2,150 440 9.0 1.08 1.33 1.24 2.0 12.2 10.9 0.67 11 KH KH Kanagawa, Higashitotsuka Hall 3,576 482 8.6 1.18 1.11 0.87 3.1 5.4 8.7 0.72 10 SWBS 1.0 KM Kirishima, Miyama Conceru 8,475 770 15.8 1.84 1.80 1.12 -0.1 8.2 8.3 0.75 26 TS PM Prague, Martine Hall 2,410 201 18.4 1.76 2.19 1.12 -1.9 12.6 12.6 0.68 11 SG Salzburg, Grossersaal in Mozarteum 4,940 844 11.5 1.66 2.06 1.07 -1.6 9.9 9.6 0.69 27 SW Salzburg, Wiennersaal in Mozarteum 1,070 209 8.4 1.11 1.33 1.09 1.7 14.9 14.3 0.77 15 0.5 TC Tokyo, Casals Hall 6,060 511 17.8 1.67 1.79 1.00 -1.3 7.6 9.4 0.71 15 5101520 TD Tokyo, Harumi Concert Hall 6,800 767 13.3 1.66 1.83 1.09 -0.1 9.8 10.8 0.71 24 V/SA (m) TH Tokyo, Hamarikyu Asahi Hall 5,800 552 14.7 1.67 1.82 0.93 0.0 7.1 8.8 0.71 15 Fig. 1 Plot of RT’s (occupied) vs. volume TI Tokyo, Ishibashi memorial Hall 5,450 662 14.9 1.70 1.84 1.10 -0.8 9.2 10.8 0.75 19 TM Tokyo, Mitaka Arts Center 5,500 625 13.3 1.73 2.28 1.02 -2.2 9.1 11.1 0.75 17 divided by the acoustical area of audience. TS Tokyo, Sumida Small Sized Hall 1,460 252 9.7 0.93 1.08 1.03 2.8 8.1 10.6 0.73 8 5 TT Tokyo, Tsuda Hall 4,500 490 12.5 1.33 1.42 0.90 0.8 7.6 10.7 0.71 20 C80,3B,occ vs. EDT VB Vienna, Brahmssaal 3,390 604 10.0 1.63 2.37 1.16 -2.8 12.8 13.6 0.77 7 4 TS VM Vienna, Mozartsaal in Konzerthaus 3,920 716 9.1 1.49 1.79 1.14 -0.2 11.6 10.8 0.70 11 BS KH SW VS Vienna, Schubertsaal in Konzerthaus 2,800 336 15.6 1.98 2.54 1.14 -3.3 14.7 13.6 0.77 12 3.5±0.5dB 3 AC ZT Zurich, Kleinersaal in Tonhalle 3,234 610 9.3 1.58 2.11 1.18 -1.8 14.1 13.2 0.70 18 Europe Japan Table 2 Correlation coefficients among acoustical factors in 18 chamber halls. 2

(dB) TT RT RT EDT C C G G IACC BR ITDG Width V N VM M M M 80,3B 80,3B L M E3 TD 80, occ 80, 1 unocc occ unocc occ C KM TH RTM, unocc - SGZT RT - 0 TI M, occ 0.93 0.1±1.6dB TMVB EDTM 0.98 0.88 - TC -1 PM C80,3B, unoccu. -0.96 -0.87 -0.98 - Bold : > 0.6

C80,3B, occ. -0.91 -0.94 -0.92 0.95 - VS

GL 0.30 0.05 0.37 -0.33 -0.12 - -2 1.0 1.5 2.0 2.5 3.0 GM 0.21 -0.05 0.32 -0.30 -0.10 0.89 - EDT (sec) IACCE3 -0.25 -0.19 -0.19 0.19 0.15 -0.07 -0.25 - Fig. 2 Plot of C80’s (occupied) vs. EDTM’s BR 0.27 0.08 0.30 -0.25 -0.07 0.80 0.56 0.11 - ITDG 0.21 0.37 0.12 -0.14 -0.23 -0.18 -0.34 0.07 -0.04 - (unoccupied). 16 Width -0.03 0.16 -0.16 0.20 0.07 -0.56 -0.65 0.20 -0.34 0.49 - SW V0.390.61 0.26 -0.29 -0.44 -0.57 -0.67 -0.08 -0.29 0.66 0.66 - GL vs. BR VS N 0.40 0.43 0.28 -0.28 -0.28 -0.29 -0.46 0.08 0.03 0.64 0.61 0.75 - 14 ZT AC 0.8 VB PM 12 BS VM 0.7 4.5dB SG

(dB) 10

L TD

G TI 0.6 TM E3 8 TS KM TT TC TH

1-IACC 0.5 6 Fig. 4 Plot of [1-IACCE,t] against the Europe 80ms KH names of halls (not rank-ordered). Japan 0.4 50ms 30ms Integration limit, t, was varied from 30 4 0.7 0.8 0.9 1.0 1.1 1.2 1.3 BR 0.3 to 80 msec. VB SW TT TM KM TS ZT BS VS TC AC TI TH SG VM TD PM KH Fig. 3 Plot of GL’s (occupied) vs. BR’s (occupied). Hall Optimum Design of a Concert Hall by Genetic Algorithms

A. Takizawaa, K. Otorib, T. Hayashib, H. Sakaib, Y. Andob, and H. Kawamuraa

aDepartment of Architecture and Civil Engineering, Kobe University, Rokkodai 1-1, Nada, Kobe, Japan bGraduate School of Science and Technology, Kobe University, Rokkodai 1-1, Nada, Kobe, Japan

Abstract: Recently, Genetic Algorithms (GA), which is one of the evolutionary computing, are applied to various complex engineering problems. An optimization system of a concert hall by employing GA with four orthogonal preferences and three models are discussed. The model 1 is that the form is based on the general shoebox type, and its proportion is optimized. The model 2 is that its plan is optimized. The model 3 is that the form is also based on the shoebox type but each wall is divided into triangles and their vertex positions are optimized. The sound simulation was performed by the image method. The results show that the optimized form of the model 1 is similar to Grosser Musikvertinssal. Those of the model 2 have different characteristics depending on the preference. Those of model 3 are various and complex ones, but they have high sound preferences.

INTRODUCTION figure1. Widths and lengths are almost same with each other, but heights show opposite characteristic. Table1 In the field of concert hall design, there is an shows the comparison of proportions between the established theory that a shoebox form has high sound results and Grosser Musikvereinssaal which is famous efficiency. On the contrary, if such simple form does as having very good sound performance. Length/width not fit to architects’ sense, circular or elliptical ones ratios are almost same. The height/width ratio of with the inevitable problem of sound focus have often Grosser is middle of the results. The subjective been used. Recently, Genetic Algorithms (GA) [1], preferences employed here seem to be appropriate for which are one of the evolutionary computing, are evaluating a concert hall. applied to various complex optimization problems. In this paper, GA are applied for a concert hall design in S:1 Lager is better. S:4 14 order to search a form having good sound performance Initial model = -0.70 Initial model = -0.30 Optimized = -0.55 free from such preconception. Optimized = -0.26 14 In the following sections, three optimization models 20 and results are discussed. Four orthogonal factors in 10 relation to subjective preference, LL, Dt1, Tsub, and IACC [2], were employed as fitness functions. The sound simulation was performed by the image method. 35m 36 The model 1 and 2 used motif-B, and model 3 used motif-A for evaluation. S FIGURE 1. Results of the model 1: (a) the result by 1 , S (b) the result by 4 . MODEL 1 AND RESULT Table 1. Comparison of proportions between the optimized forms and Grosser Musikvereinssaal. Firstly, the proportion of shoebox form is optimized. length/width height/width Width of the initial form is 20m, stage length is 12m, (a) The result by S 2.50 0.71 seats length is 30m, and height is 15m. The sound 1 (b) The result by S 2.57 1.43 source is put on the center of the stage and 72 listening 4 Grosser Musikvereinssaal 2.55 0.93 points are prepared. Each moving range of sidewalls and ceiling is ±5 m from the initial form, and each moving length of them is coded on a chromosome of

GA. Two values of S 1 and S 4 which are averaged MODEL 2 AND RESULT subjective preferences of LL and IACC by all listening points are employed. Next, the initial form of the model 1 is changed a little.

The results of optimization by S 1 and S 4 are shown in Front and rear walls are divided vertically to two ones, and each sidewall is divided to 5 ones. The coordinates and evaluation values. Figure 4 shows the optimized of two bottom vertexes of each surface are result of md2. Center of the each wall except for the parameterized. The moving range of each vertex is ceiling is swelling outside. The ceiling is folded along ±2 m in the direction of the surface’s normal line. the centerline of the hall. Two protuberances circled in Figure 2 shows the results. Front and rear walls have figure 4(b) supply sound especially to the corner of the opposite characteristic between (a) and (b). If S 1 is seats just beside the stage. considered, sounds should be reflected to seats directly.

This means the decrease of S 4 . Table 2. Each model’s connection pattern and evaluation value number ceiling sidewall front/rear wall evaluation value md1 1 1 1 -0.55 md2 1 1 2 -0.49 md3 1 2 1 -0.60 md4 1 2 2 -0.52 md5 2 1 1 -0.53 md6 2 1 2 -0.52 md7 2 2 1 -0.60 md8 2 2 2 -0.60 S =-0.20 S =-0.15 1 4 FIGURE 2. Results of the model 2: (a) the result by

S 1 , (b) the result by S 4 .

(b) MODEL 3 AND RESULT

The model 3 uses a little complex model. Shown at 1st step of figure 3, the model is consists of a ceiling, a (a) front wall, a rear wall, two sidewalls, a stage, and a (c) floor. At the 2nd step, vertexes for triangle division are FIGURE 4. The result of model 3 (md2): (a) the plotted on each surface except for the stage and floor. whole view, (b) the front view, (c) the left view. At 3rd step, each surface is divided to some triangles by connecting the vertexes. Two connection patterns are supposed. The coordinates of each vertex are CONCLUDING REMARKS parameterized and optimized. A value that four preferences are summed and averaged by 20 listening In conclusion, we would like to state the following points is used for evaluation. three points. 45 ± 10m (1) The subjective preferences used here seem to be 1st Step appropriate for evaluating a concert hall from the Ceiling Rear wall 3rd Step 29 ± 10m similarity of proportions between the optimized Side wall Connection pattern 1 form of model 1 and Grosser Musikvereinssaal. Front wall 15 ± 5m (2) There is a tradeoff on a concert form between the preference S and S . Stage Floor 1 4 (3) There could be many complex and various forms

2nd Step 1 2 3 having higher preference values than the Connection pattern 2 conventional shoebox form. 1

2 REFERENCES FIGURE 3. The optimization model 3. 1. J. H. Holland, Adaptation in Natural and Artificial Combination of each surface’s connecting pattern Systems, The University of Michigan Press (1975) produces eight different initial models. They were all 2. Y. Ando, Architectural Acoustics, Springer-Verlag New used for GA optimization. Table 2 shows their details York (1998)

Effects of Scattered Reflections by Array of Columns Measured after Construction of the “Tsuyama-Music- Cultural Hall” Y. S u z u mu r a a,b and Y. Andoa

a Graduate School of Science and Technology, Kobe University, Rokkodai, Nada, Kobe, 657-8501, Japan, b Urban Design Union, Harbor Land Center Bldg. 1-3-3 Higasi-kawasaki, Chuo, Kobe, 650-0044, Japan

The acoustical design of the Tsuyama-Music-Cultural Hall was made based on the theory of subjective preference. The hall is called “Bell Forêt Tsuyama” due to a number of circular columns, realizing the similar effects of scattered reflections by trees in a forest for the sound field of this hall. The array of these circular columns is designed to obtain scattered sound field and to decrease the value of IACC in the audience seats. In order to examine the quality of the sound field, the four-orthogonal-acoustic parameters of the sound field were analyzed using the system developed based on the subjective preference theory. From the measurement of the IACC after construction, it is shown that the sound field of this concert hall is much improved by existing the array of circular columns.

INTRODUCTION in the real hall. An omni-directional loudspeaker was placed at a height of 1.2m above the center of the stage The purpose of this work is to show that the sound as the sound source. Sound signals were recorded field in this concert hall is improved by the array of 52 through two microphones at ears entrance of a real circular columns (diameter: 30 cm) installed in front of head at 15 seat positions. After obtaining the impulse the walls both in the audience area and in the stage response, four-orthogonal-acoustic parameters were enclosure. Effects of these columns on the sound field analyzed, and scale values of the subjective preference in the audience area have been discussed reconfirmed were calculated. These four orthogonal acoustic by a previous study using the 1/10 scale model [4,5]. parameters are listening level (LL), initial time delay We described in the study that values of IACC between the direct sound and the first reflection ( T1), decreased and the initial time delay gap was prolonged subsequent reverberation time (Tsub), and magnitude due to the effects of the columns. To calculate the of the inter-aural cross-correlation (IACC). FIGURE 1 effects of scattered reflection on sound field is shows the plan of Bell Forêt Tsuyama with the array of extremely laborious, in this reason, we adopted the circular columns and the 15 measurement points. experimental method to evaluate the sound field involving scattered reflections. 13 14 15 9 10 11 12 PROCEDURE 5 6 7 8 1 2 3 4 Loudspeaker The measurement after construction was made under the similar condition to 1/10 scale model experiment Columns previously performed [4,5]. Unfortunately, we could not make the measurement without the columns array Column 5m Measurement points FIGURE 1: Plan of the Concert Hall and Measurement Points Table 1. Values of IACC Calculated by use of Architectural Scheme and Measured in the Real Hall Calculated and Measured 15 seating Position 123456789101112131415 Simulation without 0.52 0.24 0.28 0.28 0.57 0.36 0.27 0.24 0.58 0.52 0.34 0.22 0.45 0.56 0.26 Columns and Reflectors at Real Hall with Columns 0.41 0.34 0.36 0.19 0.26 0.40 0.28 0.25 0.15 0.30 0.27 0.28 0.21 0.16 0.20 without reflectors at 500

Simulation without 0.45 0.16 0.25 0.20 0.51 0.27 0.18 0.10 0.44 0.35 0.21 0.24 0.28 0.51 0.31 Columns and Reflectors at Real Hall with Columns 0.31 0.18 0.14 0.11 0.08 0.23 0.13 0.13 0.23 0.15 0.08 0.12 0.26 0.23 0.09 without reflectors at 1000

Simulation without 0.19 0.31 0.18 0.08 0.26 0.24 0.09 0.20 0.30 0.23 0.18 0.23 0.33 0.47 0.37 Columns and Reflectors at Real Hall with Columns 0.13 0.14 0.14 0.11 0.13 0.09 0.13 0.09 0.14 0.07 0.14 0.07 0.09 0.09 0.08 without reflectors at 2000

RESULTS scattered reflections above 1000 Hz. Columns array has large effects on the quality of the sound field in a Table 1 compares results of the simulation by the use concert hall and the values of IACC at the seats near of architectural scheme (without columns) and the the side walls become small by existing the array of the measurement in the real hall (with columns). These circular columns. It should be concluded, from what comparisons can be summarized as follows: has been clarified in this measurement in the real hall, 1. Values of IACC decrease in side area near that these phenomena may be caused by the scattered the sidewalls (No.9 – 15) more than the center area reflections of the columns array. The array of the (No.1 – 8). circular columns is effective on the values of IACC, 2. The maximum value of IACC appears, in the especially at above 1000 Hz, and thus improves the real hall, at the center area near the stage at 500 preference of the sound field in this concert hall. Hz. 3. Measured results shows that values of IACC ACKNOWLDGEMENTS become small as the frequency increases. And, number of audience seats obtaining smaller IACC The authors would like to thank I. Yamamoto, and T. increases with increasing frequency due to Iizuka for their measurement works of this hall and columns. These results may be as a typical Masanao Ohwaki for his cooperation. scattering effect by columns. REFERENCES DISCUSSION AND COCLUSION 1. Y. Ando, Concert Hall Acoustics, Springer-Verlag, Berlin (1985). 2. Y. Ando, Architectural Acoustics, Blending Sound Sources, Sound The acoustical design of this hall was made at three Fields and Listeners, AIP Press/Springer-Verlag, New York steps based on the theory of subjective preference. The (1998). 3. H. Sakai, S. Sato, and Y. Ando, J. Acoust. Soc. Am., 104, 1491- first is the basic shape planning based on the theory of 1497(1998). subjective preference, the second is the case study of 4. Y. Suzumura, Y. Ando, M. Oowaki, T. Iizuka, and I. Yamamoto, the shape of this hall using a computer simulation Forum Acusticum, Berlin (1999). system, and the third step is the study about the effects 5. Y. Suzumura, M. Sakurai, Y. Ando, I. Yamamoto, T. Iizuka, and M. Oowaki, J. Sound Vib. 232, 303-308 (2000) of the columns array using the scale model of this hall [5]. It has shown in the third study that the diameter of the circular column is effective on the frequency of

Blending Architectural and Acoustic Factors in Designing an Event-hall

A. Takatsua, H. Sakaib, and Y. Andob

aShowa Sekkei Co., 1-2-1-800 Benten, Minato-ku, Osaka 552-0007, Japan bGraduate School of Science and Technology, Kobe University, Rokkodai, Nada, Kobe 657-8501, Japan

To blend architectural design with acoustic design, a design-process consisting of temporal and spatial factors is proposed. As an application of this design-process, a multi-purpose-event-hall, which is the part of complex-architecture, is demonstrated. To examine the sound field, acoustic measurement was conducted to obtain temporal and spatial factors in a sound field after construction. One goal of this project was to solve acoustic problems caused by the round shape of the event hall, where the architectural design was previously determined by a certain competition of the complex, in which the architectural concept and theme was proposed. Nevertheless, the acoustic problems have been solved without unduly affecting the architecture of the hall, and this process would have been considered to be successful. In addition, some knowledge of methods to solve acoustic problems, caused by the round shaped architecture, was obtained through the designing with the process blending architectural and acoustic factors.

REQUIRED CONDITIONS INTRODUCTION ・�requirement of customers ・�social condition A process of designing halls and theatres, in which ・�natural condition the temporal and spatial design of architecture is demonstrated by the temporal and spatial factors of ARCHITECTURAL CONCEPT acoustics, is proposed (Fig. 1). A round-shaped multi-purpose event hall, the ORBIS Hall (Fig. 2) in a ACOUSTIC CONCEPT CONCEPT complex (Kobe Fashion Plaza), was designed using this process. The sound field was measured to examine acoustic factors [1,2] after construction. UTILIZATION PLAN FLOW PLANNING SECTION PREFERENCE AESTHETICS, PROPORTION BLENDING ARCHITECTURAL AND ARCHITECTURAL DESIGN

ACOUSTIC FACTORS (1) (3) (2)

Both of architectural design and acoustic design BLENDING ARCHITECTURAL AND ACOUSTIC DESIGN

were processed by temporal design and spatial factors. ���t1 IACC In order to blend architectural design and acoustic Tsub SPL ACOUSTIC DESIGN design, it is necessary to consider blending temporal and spatial design of architecture with temporal and TEMPORAL DESIGN SPATIAL DESIGN spatial design of acoustics mutually as shown in Fig. 1. (1) Blending the temporal factors of architectural FIGURE 1. Design process blending architectural design design with those of acoustic design and acoustic design. In order to control appropriate Tsub for any kind of events, a hybrid-reverberation control system was adopted. The subsequent reverberation time Tsub of this (3) Blending the temporal and spatial factors of hall is initially designed for speech. The target value at architectural design with those of acoustic design 500 Hz was 0.7 s in the designing stage. In line with Under-floor space was taken into consideration in this, in order to accommodate not only speech but also designing a sound field at each seat, since the sound events of acoustic sound, an additional system field below the ears is equally important as well as that enhancing subsequent reverberation, which consists of above ears. To eliminate the SPL-dip in the low a reverberation-control-room and an electrical acoustic frequency range, sound path to under-floor space, system, was designed. which is considered as one of the temporal factors (2) Blending the spatial factors of architectural design controlled by architectural design, was effective [2]. In with those of acoustic design the area in front of end-stage, a perforated floor with Various equipment and devices including reflectors 5-mm diameter holes in its grid of 15-mm was were designed to improve IACC and get uniformity of designed in order to fuse above- and under-floor space. SPL in the seat area. The seating areas to the side and in the back have

movable chairs that can be stored into the under-floor (2) Through the spatial design both of architectural as convenient storage. At the steel plates of the chair and acoustic design, the reflector panels in the side of basement under the chair legs, there are drilled holes stage and seating-area are clarified to be efficient to of a 25% ratio, to the extent that strength permits. This decrease IACC and to get uniformity of SPL [4]. also allows sound waves to pass through to the (3) The room projected at the rear-end of the hall, under-floor space, eliminating the dip of which has 4.0 m in width and 3.0 m in depth, is low-frequency-range. In addition, a room for mother effective to eliminate echo disturbance “whispering and baby directly facing the end-stage was designed to gallery effect.” prevent echo-disturbance, “whispering gallery effect” (4) Through both temporal acoustic design and spatial [3]. architectural design, the SPL-dip in the low-frequency due to the reflection from the floor improved effectively by the perforated floor (Fig. 3) [5].

10 meeting� room room dressingroom (a)

guest room 0

piano-storage back stage -10 opening/closing� reflector officeticket office -20 6 m cloak room 8 m reversible reflecor 10 m seating area atrium -30 foyer 10 (b) Relative SPL [dB] foyer room for mothers� 0 and children bar corner -10 center diffusion panel reflecting panel above� end-stage -20

small diffusion panel reversible reflector -30 0 200 400 600 800 1000 underfloor space Frequency [Hz] heavy-bass speaker

FIGURE 3. Relative SPL as a function of frequency up to FIGURE 2. Plan and section of ORBIS Hall with various 1kHz. (a): Relative SPL on the perforated floor; and (b): acoustic equipment. Relative SPL on the hard floor.

MEASURED RESULTS AND REFERENCE CONCLUSIONS 1. Ando, Y ., Concert Hall Acoustics, Springer-Verlag, After construction, acoustic measurement was Heidelberg (1985). performed. In the results, the various problems, which 2.Ando, Y., Architectural Acoustics, Blending Sound Sources, are usually occurred by the round-shaped form, were Sound Fields, and Listeners, Springer-Verlag/ AIP Press, excluded. It is thought that efficiency of the proposed New York (1998). design-flow was verified, because acoustic problems 3.Takatsu, A., Sakai, H., and Ando, Y., Journal of Building could be solved without breaking the architectural Acoustics 7(2), 113-125 (2000). concept under such a worst acoustic condition. 4.Takatsu, A., Mori, Y., and Ando, Y., “The architectural and Efficiencies of the various equipment and methods acoustic design of a circular event hall in Kobe Fashion of acoustic to eliminate acoustic problems of round Plaza”, in Music and Concert Hall Acoustics, Conference shaped hall, are as follows. Proceedings from MCHA 1995, edited by Y. Ando and D. (1) Through the temporal design both of architecture Noson, Academic Press, London, 1998, Chapter 30. and acoustic design, efficiency of the hybrid- 5.Takatsu, A., Hase, S., Sakai, H., Sato, S., and Ando, Y., J. reverberation control system, which consists of Sound Vibration 232(1), 263-273 (2000). architectural- and electric-acoustic, was verified in multi- purpose event-hall.

The Acoustical Renovation of the Palais des Beaux-Arts Concert Hall in Brussels

D. Commins

commins acoustics workshop, 15, rue Laurence Savart, F-75020 Paris, France, [email protected]

Originally, the acoustics of Salle Henry-Leboeuf in Brussels was renowned. Over the years, poor maintenance and clumsy renovations contributed to the deterioration of its acoustics and aesthetics. In the 1990’s, measurements were performed and an extensive investigation of Horta’s archives, notes and drawings, was conducted. Most of the details built by Horta were then explained and, on this basis, a new renovation programme was decided with, as main goal, the restoration of the original acoustics of the hall. According to users and audiences, the original acoustics seem to have been recovered.

INTRODUCTION The main Belgian concert hall, the so-called Salle Henry le Boeuf of the Brussels Palais des Beaux-Arts has been inaugurated on October 19, 1929. At the time, it was considered to be one of the very best concert halls in the world[1].

The concept and the actual detailed design were led entirely by the architect himself, Victor Horta, a key figure of the Art Nouveau school.

Enquiries conducted in 1945 by F. Winckel and around 1960 by L. Beranek, in particular by questioning major orchestra conductors, has confirmed the reputation of this concert hall: the Salle FIGURE 1: Original stage cross-section by Horta. Henry le Boeuf was rated then at the level of the Grosser Muzikvereinsaal in Vienna, the The main parameters are as follows: number of seats Concertgebouw in Amsterdam and Symphony Hall in 2 NA : 2150, public area SA = 1300 m ,stage So = 186 Boston. 2 2 m , total area ST = 1486 m , V / SA = 8.4 or 9.4 according to various estimates, V / NA = 5.8 or 6.5 The Palais des Beaux-Arts concert hall was famous according to various estimates, SA / NA = 0,60, height for its rich bass response, its intimacy and its warmth of the stage: 92 cm above main floor, first row. and for enhancing the sound of the violin. This particular characteristic is of importance since, in The materials were as of May 1997: ceiling: 75 % those days, the Belgian school of violin was plaster on metal grid, 20 % in heavy glass on heavy considered, with Moscow, to be the best. metal structures, damped by a wire mesh (cf. Horta), 5 % of light systems; walls: plaster on brick residue, Over the years, the hall has been transformed and new painted; columns: plaster on concrete; main floor: pine technology has been introduced. Its acoustics on 75 mm sleepers on concrete; upper floors: pine deteriorated: it became dry and lost its extraordinary glued directly on concrete; stage floor: wooden floor bass qualities. The complex wooden stage was on concrete floor ( under the concrete floor Horta replaced by a concrete box. designed a large resonant cavity; originally the floor was pine with oak veneer as top layer); carpeting: A DESCRIPTION OF SALLE HENRY thick carpet on foam in the stalls, balcony and boxes LE BOEUF (the original carpet was presumably thin or non- existent); seats of the stalls, dress-circle, balcony and The cross-section of figure 1 shows some details boxes: absorption on all sides, thick seat and back (not introduced by Horta, including a genuine resonant the original); galleries: upholstered seats, thick wood chamber under the stage. layer under the seat and thin wood layer on the back. AN INVENTORY OF KNOWN MODIFICATIONS Numerous changes have taken place: an absorptive carpet has been installed; the seats have been replaced several times; the original orchestra wooden stage has been destroyed in the early seventies and replaced by a concrete stage with a wooden floor on thin sleepers; a makeshift orchestra pit has been introduced, probably around 1975, in an attempt to make the room multipurpose; the room has been painted and even redecorated several times; the original organ, which did not seem to be a success when it was inaugurated in early November 1930, has been destroyed; lights and other electrical equipment have been modified several times and many openings have been made in the ceiling for cables and lights.

THE 1999 RENOVATION FIGURE 2. View towards the stage

Extensive acoustical measurements have been 3 RTBF 1961 performed and a thorough investigation of Horta’s 2,5 archives has been conducted before renovation. 2 Raes 1961 Measurements before renovation 1,5 After Most of the MLS tests were performed in the empty 1 renovation

hall but also with a full audience[2]. Time (Seconds) Before 0,5 The hall was found to be quite dry. It is partly due to renovation the relatively small volume but it is also the 0 consequence of various low, medium and high 125 250 500 1000 2000 4000 8000 frequency absorption mechanisms that did not exist in the original design. The room impulse response was Frequency close to the typical response expected from a good concert hall of elliptical shape. FIGURE 3. Measurements before and after The renovation CONCLUSIONS

From this data, a very careful renovation was planned Careless renovations of concert halls and opera houses by Architect Georges Baines in an attempt to recover may considerably alter the acoustics. The example of the original Art Nouveau aspect of the hall and its the Salle Henry le Boeuf shows that it may be possible original acoustical qualities. The key elements were: to recover most of the original features. the reconstruction of a genuine wooden orchestra stage with a resonant cavity, a genuine wooden floor The author wishes to thank the Palais des Beaux-Arts on sleepers in the stalls, the elimination of openings of and the Horta Museum for giving him the opportunity various nature, indirect air intake and exhaust, to analyse this problem. Special thanks are due to Prof. acoustical insulation. G. Vermeir for positive contributions during the construction phase. Measurements after renovation REFERENCES

The measurements performed after completion of the 1 L. Beranek, How they sound, concert and opera halls, Woodbury: renovation demonstrate that most of the original Acoustical Society of America, 1996, pp. 189-192. acoustical characteristics have been recovered. 2. D. Commins, Proc. Institute of Acoustics, 19, 213-220 (1997). Dissimilarity Judgments in Relation to Temporal and Spatial Factors for the Sound Field in an Existing Hall

Takuya Hotehama, Shin-ichi Sato and Yoichi Ando

Graduate School of Science and Technology, Kobe University, Rokkodai, Nada, Kobe 657-8501, Japan

To examine the relationships between the subjective attribute and physical factors of sound fields, dissimilarity judgments for different source locations on the stage were performed. This study is based on the model of the auditory-brain system, which consists of the autocorrelation and crosscorrelation mechanisms for sound signals arriving at two ears and specialization of human hemispheres. There are three temporal factors (τ1, φ1, τe) extracted from the autocorrelation function and four spatial factors (LL, IACC, τIACC, WIACC) from interaural crosscorrelation function of binaural signals. In addition to these temporal and spatial factors, the orthogonal factors of the subjective preference for the sound field were taken into account. The relationships of the scale value of dissimilarity and these acoustical factors were analyzed by means of the multiple regression analysis. The results show that the calculated scale value of dissimilarity agrees with the measured scale value.

INTRODUCTION PROCEDURE

A theory of primary sensations and spatial Dissimilarity judgments were performed in the sensations to environmental noise that is based on the "ORBIS Hall" with 400 seats (Figure 1). An anechoic model of the auditory-brain system was previously source of orchestra music ("Water Music" Suite No.2 - proposed [1, 2]. Primary sensations -loudness, pitch, Alla Hornpipe by Handel) was used as a source signal. timbre and temporal duration- and spatial sensations Six loudspeakers were placed on the stage. Twenty can be described by temporal and spatial factors listeners were divided into four groups and seated at extracted from the autocorrelation function (ACF) and the specific positions. Without moving seat to seat, the interaural crosscorrelation function (IACF) dissimilarity judgments were performed while respectively. From the ACF analysis, effective switching the source locations to obtain a scale value duration of the envelope of the normalized ACF (τe), of dissimilarity. The listeners were asked to judge the the delay time of the first peak (τ1), and its amplitude subjective difference between the paired stimuli on a (φ1) were extracted. From the IACF analysis, the scale that have opposite ends: "not different" and listening level (LL), IACC, interaural delay time at "extremely different ". The judgment consisted of fifteen pairs that is the possible combinations of six which the IACC is defined (τIACC) and width of the sound fields at each listener's location. The duration of IACF at the τIACC (WIACC) were extracted. It has been shown that the environmental noises can be the source signal was 4 s, and the silent interval characterized by these factors [3, 4]. The speech between stimuli was 1 s. Each pair of sound fields was intelligibility of spoken syllable and the delay time of separeted by an interval of 5 s, and the pairs are a single reflection of sound fields can be calculated by arranged in random order. This session was repeated temporal factors extracted from the ACF [5, 6]. In five times. concert hall acoustics, the theory of subjective preference allows us to calculate the scale values of subjective preference in terms of four orthogonal factors as follows: LL, the initial time-delay gap between the direct and the first reflection (∆t1), the subsequent reverberation time (Tsub) and IACC [1]. In this study, dissimilarity judgments for different source locations on the stage in an existing hall were performed in order to examine relationships between the subjective attribute and the physical factors based on the auditory-brain system of sound fields and the theory of subjective preference by means of FIGURE 1. multivariate analysis. Plan of the "ORBIS Hall". A~D: listeners’ locations. 1~6: source locations. MULTIPLE REGRESSION ANALYSIS The relationship between the scale value obtained by dissimilarity judgments and the calculated In order to examine the relationship between the dissimilarity at each group of four seats is shown in psychological distance and physical factors obtained Figure 2. The correlation coefficient was 0.85 (p < by acoustical measurements, the data were analyzed by 0.001). Results show that the psychological distance the multiple regression analysis. For the explanatory can be well described by the factors based on the variables, a distance between paired stimuli was auditory-brain system and the subjective preference introduced by applying the factors extracted from the theory. running ACF and the running IACF analysis of recorded sound signals. In this analysis, the acoustical TABLE 2. The partial correlation coefficients. factors in relation to subjective preference were included in the explanatory variables, because the property of sound fields must be taken into account [1]. The explanatory variables were: (1) DLL, (2) Dτ1, (3)

Dφ1, (4) DIACC, (5) DτIACC, (6) DWIACC, (7) D∆t1 and (8) 3.0

DTsub. In order to construct scale value of dissimilarity among sound stimuli for the dependent variable, the original data obtained by dissimilarity judgment were 2.0 categorized to seven categories, and a method of successive categories was applied to the categorized data. Correlation coefficients among explanatory variables were examined (Table 1). The results showed 1.0 that the DWIACC highly correlated with the DτIACC. To Scale value of dissimilarity avoid the effect of multicollinearity, the DWIACC, which less correlated with the dependent variable than the 0.0 0.0 1.0 2.0 3.0 DτIACC, were also eliminated from them. In the multiple regression analysis, the distances Calculated dissimilarity for factors were combined linearly due to the expression given by FIGURE 2. Relationship between the distance scale obtained by dissimilarity judgments and the calculated τ φ τ ∆ D = aDLL+bD 1+cD 1+dDIACC+eD IACC+fD t1+gDTsub distance scale. (1) ACKNOWLEDGMENTS where a, b, c, d, e, f and g are the coefficients to be evaluated. The coefficients were obtained by a step- The authors wish to thank to the stuff of our wise regression method. laboratory for their cooperation during experiments. We also thank to the students who participated in the TABLE 1. Correlation coefficients among explanatory experimental sessions. variables. REFERENCES

1. Ando, Y., Architectural Acoustics -Blending Sound Sources, Sound Fields, and Listeners-, AIP/Springer- Verlag, New York, 1998. 2. Ando, Y., Journal of Sound and Vibration, 241, 3-18 (2000). 3. Sakai, H., Sato, S., and Prodi, N., and Pompoli, R., Journal of Sound and Vibration, 241, 57-68 (2000). RESULTS AND REMARKS 4. Fujii, K., Soeta, Y., and Ando, Y., Journal of Sound and Vibration, 241, 69-78 (2000). By applying the multiple regression analysis to the 5. Ando, Y., Sato, S., and Sakai, H., "Fundamental dependent variables and the explanatory variables, subjective attributes of sound fields based on the model regression coefficients were obtained. The partial of auditory-brain system" Computational Acoustics in correlation coefficients indicated that the effect of Architecture, edited by J.J. Sendra, WIT, Southampton,

DτIACC was maximum among all. The Dφ1 and the D∆t1 1999, Chapter 4. also contributed to the dissimilarity significantly. 6. Ando, Y., Sakai, H., and Sato, S., Journal of Sound and Vibration, 232, 101-127 (2000).

Individual Differences on Subjective Preference Judgments in Relation to the Initial Time-Delay Gap of Sound Fields

Soichiro Kurokia, Masumi Hamadab , Yoichi Andoc and Hiroyuki Sakai c

aFaculty of Environmental Engineering, The University of Kitakyushu, Hibikino, Kitakyushu 808-0135, Japan bFaculty of Engineering, Kagoshima University, Korimoto, Kagoshima 890-0065, Japan cGraduate School of Science and Technology, Kobe University, Rokkodai, Nada, Kobe 657-8501, Japan

Abstract: In this study, subjective preference tests of sound fields were conducted by varying orthogonal factors of the sound field. Ten music motifs with different effective duration of the autocorrelation function (ACF) were used. The results of the most-preferred physical factors of each listener have varied widely. An attempt to clarify a relation between non-physical factors of individual difference and the preferred initial time-delay gap is made.

INTRODUCTION (Table1). Next, the preferred initial time-delay gap ∆ τ ∆ ([ t1]p) varies with variation of e. [ t1]p s were The scale value of subjective preference was recalculated by using new τe and Eqn (1) showing x evaluated by 4 orthogonal physical factors (the coordinate of preference curve. listening level (LL), the subsequent reverberation time ∆t (T ), the initial time-delay gap (∆t ) and the interaural x = log 1 (1) sub 1 [∆t ] crosscorrelation (IACC))[1,2,3]. Subjective preference 1 p in relation to ∆t1 depends on the effective duration (τe) of the ACF of a sound source. The value of τe is the PSYCHOLOGICAL EXPERIMENTS delay time at which amplitude of normalized ACF becomes 0.1 after the first delay. In this study, value of A paired-comparison test was conducted for the total

τe of music motif was measured and psychological of 33 pairs for every parameter. ∆t1 and Tsub were set experiments about non-physical factors influencing the around inferred-theory value from τe[3]. The direct individual differences are conducted. sound and the reflection were reproduced from surrounding 16 speakers in the room. Subjects were MEASUREMENT OF EFFECTIVE required to report their preference selecting one of two DURATION OF MUSIC MOTIFS sound fields. Table2. Levels of parameter for each motif Psychological experiments were conducted in the sound-field-simulation room in the Kirishima LL (dBA) 70 75 80 85 90 ∆t (ms) 1/4τ 1/2τ τ 2τ 4τ International Music Hall (Kagoshima, Japan). Ten 1 p p p p p τ τ τ τ τ music motifs with their duration of about 6 seconds Tsub (s) 1/4*(23 e) 1/2*(23 e) 23 e2*(23e) 4*(23 e) were analyzed. Dry sources were reproduced from a IACC 0.4 0.75 1 τ τ frontal loudspeaker in the room and picked up by the p=(1-log10A) e,A=4;total amplitudeof reflection microphone at the listening position. After passing through A weighting filter, ACF was calculated. First, ∆ the running τe values were calculated with the duration FACTORS INFLUENCING TO [∆t1]p 2T=2s, and the minimum running τe values were ∆ recalculated with the duration 2T equal 30 times τe Cumulative frequency of [ t1]p (Figure1) shows that music motifs were divided into 3 groups; motifs Table1. Sound motifs preferred theory value neighborhood (motif E,C), ∆ ∆ Motif Name of motif te (ms) calculated [ t1]p motifs preferred shorter t1 than theory value (motif A Royal Pavane 26 10 L,J,K,B,A,I), motifs hardly preferred theory value B Sinfonietta, Opus 48, Ⅳ 17 7 neighborhood (motif G,H). C Piano, Classical Mood 73 29 E Water Music' Suite Ⅳ 40 16 We gave subjects a questionnaire about their G Marriage of Figaro Overture 56 22 specifics, i.e. inborn, music experience, living style H Flute, Classical Mood 67 27 ∆ I Cello, Solo 29 12 and so on. In this report, the result of [ t1]p obtained J Violin, Solo 12 5 were compared in relation to the following 6 items ; K Clarinet, Solo 14 6 gender, age, musical experience, age when he or she LTrumpet, Solo 11 4 started musical activity , term of musical activity and

LJ K B A IEHC G JK 100 I L C 100 60 E

（％） B G H 40 A 50 10 t1]p (ms) Δ

[ 20 The number of the subjects the of number The frequency Cumulative

1 0 0 0.0–9.9 20.0–29.9 40.0–49.9 60.0–69.9 10.0–09.9 30.0–39.9 50.0–59.9 70.0– 10 20 30 40 50 60 70 80 [Δt1]p (ms) τe (ms) YES ∆ NO Figure1. Cumulative frequency of [ t1]p Figure2. Distribution of [∆t1]p on musical musical activity at present. Each motif was compared experience with those items. Table4 shows number of subjects in each item. As the result of analysis of variance REFERENCES (ANOVA), significant difference was obtained on some items as indicated in Table3. As regards musical experience, it relates to 1. Ando, Y., Concert Hall Acoustics, Springer-Verlag individual differences, because three motifs among 10 Tokyo, 1987. indicated a significance difference. Distribution of 2. Ando, Y., ARCHITECTUAL ACOUSTICS Blending Sound Sources, Sound Fields, and Listeners, [∆t1]p on musical experience (Figure2) shows that subjects who had musical experience tend to prefer Springer-Verlag Tokyo, 1998. 3. Sakurai, M., Korenaga, Y. and Ando, Y., “A sound shorter ∆t than subjects who had no musical 1 simulation system for seat selection”, Music & Concert experience. Hall Acoustics, Conference Proceeding from MCHA

1995, Academic Press, London, pp.51-59, 1996.

Table3. Result of ANOVA Items Motif LJKBAIEGHCAll Gender Male vs. Female 0.203 0.928 0.432 0.600 0.620 0.821 0.298 0.487 0.074 0.661 0.187 Age Under 20 vs. 20s 0.135 0.999 0.623 0.006 ** 0.746 0.344 0.412 0.154 0.752 0.011 * Under 20 vs. 0.903 0.269 0.010 ** 0.594 0.999 0.929 0.119 0.736 Over 30 20s vs. Over 30 0.459 0.150 0.026 * 0.415 0.437 0.474 0.366 0.430 0.104 Musical experience 0.227 0.757 0.004 ** 0.560 0.733 0.034 * 0.749 0.073 0.007 ** 0.836 0.021 * Starting age of Under 10 vs. 0.638 0.720 0.013 * 0.807 0.675 0.051 0.547 0.894 0.158 musical activity Over 10 Term of musical Under 10 vs. 0.890 0.719 0.244 0.232 0.006 ** 0.634 0.001 ** 0.042 * 0.761 0.072 activity Over 10 Musical activity at 0.297 0.960 0.005 ** 0.675 0.751 0.972 0.439 0.378 present 0.05>p>0.01;*,p<0.01;** Table4. Number of subjects in each item Items MotifLJ KBA I EGHCAll Gender Male 18 11 34 32 16 46 19 19 26 12 233 Female 24 14 15 15 8 35 14 15 18 17 175 Age Under 201242291607101577 20s 271647419 5529162414278 Over 303404294115042 Musical experience Yes 31 14 19 32 16 53 24 21 27 25 262 No11102915426913164137 Starting age of musical activity Under 109 0 01411698 6 770 Over 107 0 01642313107383 Term of musical activity Under 10 5 9 10 3 0 12 6 2 9 4 60 Over 10 7 6 10 9 0 14 9 7 11 6 79 Musical activity at presentYes 218 6 22163113161418165 No70080129410757 All Subjects 42 25 49 47 24 81 33 34 44 29 408

Effects of Modulated Delay Time of Reflection on the Autocorrelation Function and Perception of the Echo Junko Atagia, Reinhard Weberb and Volker Mellertb

aGraduate School of Science and Technology, Kobe University, Rokkodai, Nada, Kobe 657-8501 Japan bAG Akustik, Physik, University of Oldenburg, D-26111, Oldenburg, Germany

The indoor sound field can be regarded more or less as time-variant system because of air current and other movements. The fluctuation of the sound-pressure-level caused by an air conditioner was measured in a gymnasium. From the statistical analysis of the sound-pressure-level fluctuation, a time-variant model of the impulse response was proposed. The impulse-response model consists of the direct sound and reflections with modulated delay time. In this paper, the physical characteristics of the time- variant sound field with the modulated delay time of reflection is analyzed by the autocorrelation function (ACF) of the sound field in order to compare with the time-invariant sound field with the fixed delay time of reflection. Psychological tests on perception of the echo were conducted changing the modulation interval. The relationship between the ACF and the perception of the echo is discussed.

INTRODUCTION Direct Reflection sound sound

Previously, Ueda and Ando reported that the A0 A
fluctuation of sound pressure level (SPL) was observed in a real room, which was caused by an air conditioner

[1]. The SPL was measured in a gymnasium using pure litude tones as sound sources, and clearly fluctuates
Amp (A0 = A ) especially at higher frequencies when the air conditioner is on. Thus, an indoor sound field can be regarded as a time-variant system. Not only large room like a gymnasium but also any indoor sound field includes such a temporal fluctuation in the physical 0 Time
t 1 environment. Based on the statistical analysis of the SPL
fluctuation, a time-variant model of the impulse response was proposed. In the model, the impulse FIGURE 1. A time-variant model of the impulse response response is represented by a direct sound and including the definitions of the delay time of reflection (
t1) reflections with carrying delay tike. A simulation using and the modulation interval of delay time of reflection (
). the model confirmed the SPL fluctuation at higher frequencies as the measured data when the air conditioner was on. impulse response was used to reproduce the time- In this paper, physical characteristics of such a time- variant sound field (Figure 1). The delay time of the
variant sound field are examined by analyzing the reflection ( t1) for the time-variant sound field was autocorrelation function of the sound field with the controlled by a sinusoidal wave. The frequency of the modulated delay time of reflection. Also, concerning sinusoidal wave was set to 0.2 Hz, because it was the subjective evaluation of the sound field variation, observed to be 0.1-0.2 Hz from the analysis of the SPL the perception of the echo in the time-variant sound fluctuation in the real room [1]. The
t1 is fixed at 240 field is investigated. ms. The values of the modulation interval (
) were determined by a preliminary experiment on the just PROCEDURE noticeable difference (jnd) of the modulation of the delay time of reflection. It was performed with ten subjects. As a result, the average
value of the jnd for A musical motif was used as sound source (Royal all the subjects was 10.6 ms. Therefore, the
values Pavane by Gibbons, 10 sec). A sound field consists of are set at 0 (fixed delay time), 5.3 (1/2
jnd), 10.6 a direct sound and a single reflection as representative (
), 21.2 (2
) and 42.4 (4
) ms. of a set of reflections. The time-variant model of the jnd jnd jnd THE ACF ACALYSES 1 1.2

To examine the physical characteristics of the time- 1.1 variant sound field, the autocorrelation function is 0.5 analyzed [2]. In Figure 2, envelopes of the normalized 1 ACF () of the sound source and the sound fields are 0.9 shown. By further examining the ACF envelopes, it is

0 (240) found that the ACF envelope of the sound field with p+1

0.8 E the fixed delay time has a peak at 240 ms, which corresponds to the
t , and that it is much suppressed 0.7 1 -0.5 by adding the modulation. To represent the difference of the ACF behavior among the sound fields with 0.6 different values of
, the ratio of the ACF envelope of -1 0.5 a sound field to a sound source is defined as follows 0 5.3 10.6 21.2 42.4 and shown in Figure 3.
  [ms] p
1 ( ) E
( )
(1) p 1  () FIGURE 3. The average scale values of the perception of the p echo for all the subjects (open square) and the ratio of the   where p( ): the ACF envelope of a sound source ACF envelope of sound fields to the sound source (filled p+1(): the ACF envelope of a sound field circle) as a function of the values of
.

PERCEPTION OF ECHO DISCUSSION AND CONCLUSION

Paired-comparison test on perception of the echo was It is found that the perception of the echo becomes conducted using the same sound fields as the ACF larger with the longer modulation interval, while the analyses as the sound stimuli. The subjects judged ACF at the delay time of
t1 becomes smaller. It is which of the two sound stimuli they felt stronger echo. considered that the echo can be easily perceived The experiment was carried out ten times repetitively because the total coherence becomes lower than that of for each subject. Ten subjects aged from 22 to 27 the direct sound around the delay time of
t1 by adding participated in the experiment. the modulation. Figure 3 shows the relationship between the scale The behavior of the autocorrelation function is value of echo and the Ep+1() value at  = 240 [ms] for effective and significant to determine the physical each sound field as a function of the values of
. It is characteristics and the perception of the echo in the found that the average scale value of echo becomes time-variant sound field. significantly larger with the longer
, while the ratio Ep+1 becomes smaller. The correlation coefficient ACKNOWLEDGMENTS between the scale values and the ratio is –0.78. 0 The authors wish to express gratitude to Professor Yoichi Ando for many valuable discussions and his precious advice. Thanks also to the students who

-5 participated in the experiments for their cooperation.

REFERENCES

-10 1. Ueda, Y., and Ando, Y., “Effects of air conditioning on sound propagation in a large space”, J. Acoust. Soc. Am. 102, 2771- 2775 (1997).

-15 2. Ando, Y., Architectural Acoustics – Blending Sound Sources, 0 100 200 300 Sound Fields, and Listeners, AIP Press/Springer-Verlag, New  [ms] York, 1998.

FIGURE 2. Example of the ACF envelope; thin line: the 3. Ueda, Y., Furuichi, H., and Ando, Y., “The just noticeable sound source, circle: the sound field with fixed delay time difference in variable delay time of the single reflection”, in (
=0), and thick line: the sound field with modulated delay Music and Concert Hall Acoustics, edited by Y. Ando and D. time (
=10.6 ms). Noson, Academic Press, London, 1997, pp. 133-137. Using Artificial Intelligence to Enable Occupied Measurements of Concert Hall Parameters

Trevor J. Coxa and Francis F. Lib

a,School of Acoustics and Electronic Engineering, Salford University, Salford M5 4WT, UK b Dept of Computing and Mathematics, Manchester Metropolitan University, Manchester M1 5GD, UK

Objective room acoustic parameters are usually measured using artificial test signals, while subjective sound quality is assessed under naturalistic sound sources. Objective parameters are difficult to measure under occupied in-use conditions due to various logistical and technical problems. Many difficulties would be overcome if the naturally occurring sources in concert halls, such as music or speech, could be used. This paper presents a method to do this using an artificial intelligence engine. Artificial neural networks are trained to extract objective parameters from speech heard in the concert hall through a large database of examples. Careful pre-processing of the speech samples is required, using either an unsupervised network or an envelope spectrum estimator. From simulated speech excerpts, the reverberation time can be estimated to an accuracy of better than 0.1, with a few necessary conditions. It is noted that in low frequency subspace, speech has certain stable statistical features, and so by training artificial neural networks on a large number of different speech examples, it is likely that the network can learn to adapt to different speakers, realizing source independent measurements.

INTRODUCTION compromised accuracy when compared to using artificial signals. Inspired by the fact that human Occupied objective acoustic measurements are not hearing can sensitively differentiate reverberation often undertaken because of logistical and technical times, artificial intelligence methods were recently problems. The use of artificial test signals makes it introduced into room measurements to enable more difficult to get realistic in-use conditions, as the test accurate estimations with speech excitations [2]. In signals are unacceptable to occupants. It is suggested particular, artificial neural networks (ANNs) with that many difficulties could be overcome if the purpose designed pre-processors were developed to naturally occurring sounds in concert halls, such as extract objective parameters from speech. This paper music or speech, were used. This would give a non- will discuss these ANN methods and present some invasive test method. new techniques and results about the extraction of Such a concept is not new, for example Steeneken octave band reverberation time from running speech. and Houtgast [1] suggested estimating the Modulation Transfer Function (MTF) from running speech to ANN METHODS obtain the Speech Transmission Index (STI). The MTF was estimated by comparing the envelope spectra of The method of using supervised ANNs to extract source and received speech, but at the cost of objective parameters is depicted in Fig.1. There are two phases to developing and applying the ANN regime, namely training and retrieve. In the training Known acoustical parameters (RT) as teachers phase, the ANN “learns” to extract reverberation from speech through thousands of examples. For this, + Reverberated Neural reverberated speech examples with known acoustic Σ Speech Preprocessor Network _ parameters are used. The speech examples are pre- processed and conditioned to yield input vectors for the ANN. The output of the neural network and the Adjust NN using error signals corresponding true acoustic parameter are compared to obtain the error. The training process is to iteratively update the internal parameters of the neural network to minimize the mean square error over all training examples. In this study, a back propagation algorithm is used. The training phase is slow. The signal pre- processing plays a vital role in the systems success and is the focus of the regime’s development. For each Fig. 1 Training (top) and retrieve (bottom) phases. acoustic parameter, a different pre-processor is needed. In the retrieve phase the neural network structure is A high resolution, normalised envelope spectrum fixed. The ANN will then give accurate acoustic estimator is used as a pre-processor. This is based on parameter estimation from speech signals. This the Hilbert Transform. A better than 0.1s resolution of happens in real time, and the ANN can generalise to octave band reverberation time can be achieved in examples not previously seen during training. Data noise free cases, Fig.2. sets used to train and validate the ANN are a super set The normalised envelope spectrum contains both of reality, comprising hypothetically more than all the ambient noise and reverberation effects. This makes possible types of room impulse responses. Validation this frequency domain approach more sensitive to was performed strictly using data that had never been ambient noise than the time domain method. Although used in the training phase. Current work only includes ANNs have intrinsic capability to deal with noisy data, a very limited set of real measurements, what is this is not true in this case. To achieve a better than 0.1 presented here is mainly based on simulation. second resolution, >40dB S/N is typically required. Noise reduction techniques used in speech processing SEPARATE UTTERANCES might help, but these have not been tried. Reverberation times have been successfully extracted from separate utterances (“1, 2, 3, 4”) [2]. A set of RMS detectors monitor the slope of the rise and fall edges of these syllables - a time domain approach. Results demonstrate that the ANN can generalize. The worst cases have an accuracy of <0.1s for reverberation time. Problems arise if the utterances are not sufficiently spaced and so the reverberant decays are masked by future syllables. This method is ambient noise tolerant - ANNs usually have intrinsic noise tolerance - provided the slopes are not significantly masked by noise (say, S/N>20dB). When octave band reverberation times are estimated, however, accuracy is compromised. The problem stems from a lack of signal excitation in the octave bands. For octave band analysis, more Fig. 2 Accuracy of RT for 1kHz octave band utterances are required to ensure sufficient signal level in each band, but then the time domain pre-processor DISCUSSIONS AND CONCLUSIONS yields too many input vectors for the ANN. Consequently, a different pre-processor is needed, as ANNs can accurately estimate objective parameters outlined below. This pre-processor also has the added from speech. This provides a new method to facilitate advantage of using more natural, running speech. in-use, non-invasive measurement of concert halls. An envelope spectrum estimator coupled to an ANN gives RUNNING SPEECH octave band reverberation times for running speech. Problems arise with ambient noisy, and when arbitrary Envelope spectra are efficient pre-processors for speaker/text are considered. Further verification with running speech, as the low frequencies give more real room measurements is required. information on the changes to the speech envelopes, such as reverberation effects. Objective parameters are contained in the difference between the envelope ACKNOWLEDGMENTS spectra of the original and transmitted speech. However, the mapping between the difference of the This project was funded by EPSRC, UK (GR/L89280). envelope spectra and objective parameters are complicated and difficult to formulate. The method REFERENCES proposed in [1] only gives approximate estimations, ANNs can map the complicated functions, and so 1 H. J. M. Steeneken and T. Houtgast, Proc. 11th ICA, Vol. improve accuracy. Furthermore, explicit knowledge of 7, Paris 1983 ,P 85-88. the input signal may not be needed as the average low frequency statistics concerning pronunciation and text 2. T J Cox, F. Li and P Darlington., J.Audio.Eng.Soc. 49(4) can be learnt by the ANN. 219-230. (April 2001) Auditory Evoked Magnetic Fields Corresponding to the Subjective Preference of Sound Fields

Y. Soetaa, S. Nakagawab, M. Tonoikeb and Y. Andoa

aGraduate School of Science and Technology, Kobe University, Rokkodai, Nada, Kobe, Japan bLife Electronics Laboratory, National Institute of Advanced Industrial Science and Technology, Osaka, Japan

Previously, the relationships between the brain activities and subjective preference for sound fields have been examined by auditory evoked potential (AEP) and alpha waves range of continuous brain waves. To investigate human cortical responses corresponding to the subjective preference for speech sound field, an attempt is made here to analyze the autocorrelation function (ACF) of magnetoencephalogram (MEG) in relation to the delay time of single reflection. It is assumed that a similar repetitive feature of the
-wave range (8 – 13 Hz) of MEG in terms of the effective duration of the ACF is related to subjective preference. The source signal was the Japanese word “piano” of 0.35 s. The delay time of the first reflection (t1) was varied at five levels (0, 5, 20, 60, 100 ms). In order to compare results of MEG with the subjective preference, a reference stimulus was first presented and then an adjustable test stimulus was presented. Such a pair of stimuli was presented alternately 50 times for recording MEG in similar manner to the paired-comparison tests. It is remarkable that there is a linear relationship between the subjective preference and the effective duration of the ACF of the
-wave of MEG.

INTRODUCTION EXPERIMENTS

Four significant and independent physical The source signal was the Japanese word “piano” parameters of sound field, which takes account of time of 0.35 s. The delay time of the reflection was varied and space as they are specialized in human cerebral at five levels (0, 5, 20, 60, 100 ms). The amplitude of hemispheres, have been discovered [1]. These the reflection was the same as that of the direct sound. parameters are: (1) the level of listening, (2) the initial The sound-pressure level was fixed at 70 dBA. Seven delay gap between the direct sound and first reflection subjects, males, 23 to 26 years old, participated. The

(
t1), (3) subsequent reverberation time (Tsub), and (4) auditory stimuli were binaurally delivered through magnitude of the interaural cross-correlation (IACC). silicon tubes and earpieces. They asked to close their Recently, the effort to describe such important eyes so as to fully concentrate on the speech. qualities of sound in terms of the auditory pathways Following the paired-comparison method, each subject and brain has been brought to bear on the problem. If compared 10 pairs per session, a total of 10 sessions enough were known about how the auditory system were conducted for each subject. The interval between modifies the nerve impulse from the cochlea, the the stimuli presentations was 1.0 s, and that between design of the concert hall and opera house, for comparison pairs was 4.0 s to allow time for the example, could proceed according to guidelines subjects to respond. The subjects were asked which derived from knowledge of these processes. stimulus they preferred to hear. The scale values of the Concerning the relationship between brain subjective preference of each subject were calculated activities and subjective preferences of a sound field, a according to Case V of Thurstone’s theory [4]. method of using the autocorrelation function (ACF) Recording of magnetic fields were carried out in a was developed to analyze elecroencephalogram (EEG) magnetically shielded room using a 122-channel [2, 3]. They analyzed the effective duration of the whole-head neuromagnetometer. During recording, the envelope of the normalized ACF (e) of the alpha subjects sat in a chair with their eye closed. In order to waves when temporal factors
t1 and Tsub were varied. compare the results of the MEGs with scale values of
The results showed that e of the alpha waves is longer the subjective preference, a reference stimulus ( t1 = 0 only in the left cerebral hemisphere for the preferred ms) was first presented and then the adjustable test conditions of these temporal factors. In this study, to stimuli (
t1 = 0, 5, 20, 60, 100 ms) were presented investigate the relationship between brain magnetic using a constant interstimulus interval of 1000 ms. fields and subjective preferences of sound fields, the Such pairs of stimuli were presented alternately 50 ACF of magnetoencephalogram (MEG) is analyzed. times and the MEGs were recorded. 1.5 700 550 550 600 r = 1.00 (a) r = 0.99 (b) r = 0.94 (c) r = 0.91 (d) 1.0

0.5 600 500 500 [ms] e 0.0 500  -0.5 500 450 400 -1.0 -1.5 400 Scale value preference of value Scale 400 450 300 02040 60 80 100 02040 60 80 100 0 20406080100 02040 60 80 100


t [ms] t1 [ms] t1 [ms]
t1 [ms] 1 1.5 450 600 600 r = 0.91 (e) r = 0.90 (f) r = 0.91(g) 1.0

0.5 500 500 [ms] e

0.0 400  -0.5 400 400 -1.0

-1.5 350 300 300 Scale value preference of value Scale 0 20406080100 0 20 40 60 80 100 0 20406080100


t1 [ms] t1 [ms] t1 [ms]

FIGURE 1. Relationships between scale values of subjective preference (

) and averaged values of e obtained at a certain channel of MEG over left hemisphere (ZZ) as a function of t1 for all subjects.

To obtain a degree of similar repetitive features of than 0.87. There was a tendency, however, that the the MEG alpha waves, the effective duration of the change of e in the left hemisphere was wider than that ACF, e, defined by the delay  at which the envelope in the right hemisphere. This may indicate that the left of the ACF becomes –10 dB is determined. 16 hemisphere dominance of the human brain for such a channels of the temporal area on both left and right change of the
t1 for speech. cerebral hemispheres of each subject’s head were The auditory-brain model for describing selected to analyze. subjective attributes for sound fields, which consists of the autocorrelation and the interaural cross-correlation RESULTS AND REMARKS mechanisms, is proposed [1, 6]. In this study, the linear relationship between subjective preference and  Figure 1 shows relationships between scale values the ACF factor, e, is found. This may also imply the existence of the ACF mechanism in human brain. of subjective preference and averaged values of e over left hemisphere. The averaged values of e derived from the certain channel that showed the highest ACKNOWLEDGMENTS correlation almost directly related to the scale values of subjective preference. Thus, there is a linear This research was supported by a Research relationship between scale values of subjective Fellowship from the Japan Society for the Promotion preference and the values of e. Correlation coefficient of Science for Young Scientists. for each subject was more than 0.91 in the left hemisphere. This tendency is much more significant REFERENCES than the results of previous studies on alpha waves of EEG [2,3]. 1. Y. Ando, Architectural Acoustics, Blending Sound The left hemisphere is mainly associated with Sources, Sound Fields, and Listeners, AIP Press, speech and time-sequential identification, and the right Springer-Verlag, New York (1998). is concerned with nonverbal and spatial identifications 2. Y. Ando and C. Chen, J. Archi. Plann. Environ. [1, 5]. The response, which corresponds to subjective Engng. AIJ., 488, 67-73 (1996). preference, was only found in the left hemisphere in 3. C. Chen and Y. Ando, J. Archi. Plann. Environ. previous studies [2,3]. In this study, a linear Engng. AIJ., 488, 73-80 (1996). relationship between scale values of subjective 4. L. L. Thurstone, Psychol. Rev. 34, 273-289 (1927).  preference and the values of e are found both right 5. D. Kimura, Sci. Amer. 228, 70-78 (1973). and left cerebral hemispheres. In the right hemisphere, 6. Y. Ando, J. Sound Vib., 241, 3-18 (2001). the correlation coefficient for each subject was more Subjective Evaluations of Scattering in Rooms

R. R. Torresa, M. Kleinerb, U. P. Svenssonc, and G. Natsiopoulosd

aProgram in Architectural Acoustics, Rensselaer Polytechnic Institute, 110 8th St., Troy, NY 12180, USA bDepartment Chalmers Room Acoustics Group, Chalmers Univ. of Tech., SE-412 96 Gothenburg, Sweden cAcoustics Group, Dept. of Telecommunications, Norwegian Inst. of Technology, NO-7491 Trondheim, Norway dAkustikon, Baldersgatan 4, SE-411 02 Gothenburg, Sweden

Generic numerical parameters for the acoustical quality of concert halls (for example, reverberation time) do not adequately describe the finer perception of spectral and spatial attributes of scattering from interior surfaces. In addition, it is helpful to determine what level of computational accuracy is necessary for accurate auralization. Thus, to better understand the perception of scattering, subjective tests have been performed using various computational models of edge diffraction and surface scattering. Results from the studies show interesting relationships among the audibility and perceived quality of the sound, and the type of input signal used in the investigations.

INTRODUCTION “bosses” (e.g., hemispheres or semi-cylinders). Here we have implemented the boss model proposed in [2] The scattering behavior of faceted and rough surfaces to complement to the wedge-assemblage method, and is of particular importance in room acoustics, as it we present some initial results of listening tests when determines the spatial coverage and spectral coloration hemispherical scatterers are placed on the side walls of early reflections to listeners. For computer and ceiling of a concert-hall “shoebox” geometry. simulations, the task, then, is to choose appropriate mathematical models of surface scattering. NUMERICAL MODELING

The scattering from an embedded sphere (i.e., a Previous Subjective Studies hemisphere on a plane) is depicted in Fig. 1. The total In early studies using a simplified energy-based pressure at the receiver is the sum of the incident and Lambert “diffusion” model [1], listening tests showed reflected specular components, in addition to the that changes in the frequency-dependent diffusion incident and reflected scattered components above the coefficient were clearly audible as coloration and plane of the hemisphere. If we group the latter three spaciousness differences, where the perception of components together, we express the total pressure as these attributes relied on the input signal. Listeners simply the sum of the incident and total scattered gave entirely consistent comments regarding perceived components: 

sp
sc
sc  coloration changes in different frequency ranges. ptot pi pr pinc pr (1) For higher-accuracy modeling, a hybrid approach [2] would consist of using a wedge-assemblage method (using the Svensson et al. extension [3] of the BTM model [4]) to compute edge diffractions from pi psc room surfaces and a “boss” model to compute i sc scattering from hemispheres on wall surfaces. The pr application to a stage house has already been discussed in [5], where double-blind ABX listening tests showed that the inclusion of diffraction was audible in the psp early impulse response even for non-shadowed r receivers. These subjective tests indicated possible computational savings, e.g., the omittance of second- FIGURE 1. The total pressure is the sum of the order diffraction computations for non-shadowed incident and scattered components above the plane of source-receiver orientations. the hemisphere. The present work addresses surfaces that are not as easily represented as an assemblage of edges. Such Moreover, the method combines the method of images surfaces could be statues on walls or other isolated and an analytical solution for scattering from a rigid sphere. [The Lambert-diffusion model puts magnitude and the results show that the audibility of the boss bars around the latter three terms in Eq. 1 and scattering with early reflections depends on the input propagates the “energy” toward the receiver. Thus, the signal. It seems reasonable that the audibility of the scattered phase cannot interfere with the direct field boss scattering would increase significantly with the and thus does not yield the total field.] number and density of bosses in the hall. The test geometry is a rectangular concert hall (Fig. 2). The 12 hemispheres on each side wall and ceiling Table I. Audibility of freq. dependence in scattering. (36 total) are spaced about 6-8 m apart, which is a (Yes= significant audible difference with p < 0.5) conservative, low boss density. (Coupling between frequency Organ Stg Qtet Chirp Impulse bosses has been addressed theoretically by 125 vs 500 Yes Yes Yes Yes Natsiopoulos [6], but remains to be fully 125 vs 2000 Yes Yes Yes Yes implemented.) The diameters are one-half wavelengths 500 vs 2000 Yes Yes Yes Yes for the frequencies 125 Hz, 500 Hz, and 2000 Hz (radii 69 cm, 17 cm, and 5 cm). Table II. Audibility of scattering in early BRIR. (Yes = significant audible difference with p < 0.5) frequency Organ Stg Qtet Chirp Impulse 125 No Yes Yes Yes 500 Yes 2000 Yes S R Future work could include applying more 44 m physically based values for the scattering coefficients as input data into current auralization programs, and FIGURE 2. Test geometry for listening tests. studying whether the perceptual differences are significant between more accurate scattering models LISTENING TESTS and simpler current approaches. Descriptors for the sound quality of scattering would also be useful. Early binaural room impulse responses (BRIR) are computed. Only one order of specular reflections is used, to more easily analyze the early BRIR structure ACKNOWLEDGEMENTS and coloration, and for comparison with a Lambert Support was provided by the Johnson Foundation and model with two different implementations for first- Teknikbrostiftelsen, Sweden. order and higher-orders of reflection. The listening tests have two parts. First, the REFERENCES subjects performed an ABX test to determine whether the differences between the boss sizes were audibly 1. R. R. Torres et al., Acustica 86(6), 919-927 (2000). significant. They also rated the “general difference” on a scale and could describe the perceived difference. 2. R. R. Torres and M. Kleiner, Proc. 137th ASA/2nd EAA Four signals (an organ chord, a string quartet, a chirp, (Forum Acusticum)/DAGA, Berlin, March 1999. and the impulse itself) are used to judge their effect on the audibility. (Spectrograms and further details are 3. U. P. Svensson et al., J. Acoust. Soc. Am. 106, 2331- given in [7].) The second part compares the BRIR 2344 (1999). “with and without” bosses, to determine whether the 4. H. Medwin, J. Acoust. Soc. Am. 69, 1060-1064 (1981). difference is audible, despite the large boss spacing between them. The largest boss size is used 5. R. R. Torres et al., J. Acoust. Soc. Am. 109, 600-610 (corresponding to 125 Hz), along with the four input (2001). signals above. The total number of cases studied was necessarily limited to avoid fatigue in the test listeners. 6. G. Natsiopoulos, Report E 00-05, Applied Acoustics, Chalmers Univ. Tech., Gothenburg, Sweden (2000).

RESULTS 7. R. R. Torres, Studies of Edge Diffraction and Scattering: With 12 subjects the ABX results from the first part of Applications to Room Acoustics and Auralization, Ph.D. the listening tests show (Table I) that the frequency- thesis, Report F 00-02, Dept. Applied Acoustics, Chalmers Univ. Tech., Gothenburg, Sweden (2000). dependence is clearly audible in the scattering alone, as one should expect. In Table II, 24 trials were used, Acoustical Behavior of Small Lecture Rooms

J. A. F. Gomesa, M. R. S. Ribeirob

aInstituto Politécnico da Guarda, Guarda, Portugal bFaculdade de Engenharia, Universidade do Porto, Porto, Portugal

This paper aims to characterise the acoustical behaviour of small lecture rooms with different sizes. Characterization is accomplished by measuring room sound quality parameters such as Reverberation Time, Early Decay Time, Speech Intelligibility, Clarity, Definition, etc. The measured values are compared with those yielded by various mathematical models based on prediction techniques in order to obtain some conclusions. The rooms under study have materials with similar acoustic absorption and diffusibility features. A set of conclusions is presented in order to define the optimal features for sound quality and room acoustics prediction in small lecture .

INTRODUCTION Table 1: Lecture rooms characteristics Nº of Floor Relation Lecture Volume Ideal T places surface V/Np R Room V (m3) (s) In this study we analyze the acoustic behavior of (Np) (m2) small lecture rooms with similar finishing materials S53 337 90 112 3.7 0.75 and acoustic characteristics (absorption and S4 264 44 92 6.0 0.75 diffusibility). From this analogy we expect to evaluate S7 264 44 92 6.0 0.75 the coherence of the acoustical characteristics for each S9 265 44 93 6.0 0.75 one of the rooms and for the diverse coating materials. S23 259 35 91 7.4 0.75 All the auditoriums belonged to the same school and S25 259 35 91 7.4 0.75 they don't have any system of sound amplification. S56 167 45 57 3.7 0.70 S6 128 21 45 6.1 0.65 The parallelepipedic shape is the basic pattern. S8 128 21 45 6.1 0.65 The absorption and diffusibility features of the internal surfaces had been obtained directly from bibliographical references and from the prediction ACOUSTICAL MEASUREMENTS software data. To consider the sound absorption coefficient of the chairs, some difficulties in getting The measurements were carried out in two fases. The direct values from the literature had been verified, first one evaluate the quality of speech communication because the available information was insufficient with respect to intelligibility for the different lecture (details absence). rooms using the RASTI Index. In the second fase, first The most important architectural lecture rooms data we evaluated three 60 dB decay times: T30, T20 and are resumed in Table 1. The ideal reverberation, EDT (Early Decay Time) derived from the decays determinated by the Stephens and Bate empirical measured in the auditoriums, and then four room- method (r=4 for speech), is based in equation (1). acoustic parameters: Clarity (Early-to-late Sound From graph that relates optimum reverberation time Index), Definition (Early Energy Fraction), Centre with auditoria size, Martins da Silva [1] suggest for Time and the Total Sound Level. All these speech use, values between 0,6 and 1,0 seconds. measurements had been made for octave-bands between 125Hz and 4kHz. T  r
0 ,012 3 V
0 ,107  (1)

Another particularity from the analyzed lecture rooms PREDICTION METHOD are the high values of the relation between volume and number of seats verified in most of the rooms (6.0 - For acoustic prediction we used software based on 7.4 m3/seat). B.J. Smith and others [2] considerer as the Randomize Tail-corrected Cone-tracing (RTC) – correct volume/person for lecture rooms a value from CATT–Acoustic - developed by CATT (Computer 2,8 m3/pers. (optimum) to 4,9 m3/pers. (maximum). Aided Theatre Technique) from Gothenburg - Sweden. To use the prediction software it's necessary to define the room geometry (corner coordinates and the plane definitions) as well as the receiver's position and source data. As the author refers [3] in the user’s manual the underlying theory is the geometrical Total Sound Level (Ls), the Definition (D50) and the acoustics, the reason why for small rooms only the Reverberation Decays T15 and T30. upper octaves 1, 2 and 4 kHz will be well predicted. In For the considered Sabine sound absorption the resulting analysis we will try to verify this aspect. coefficients of coating materials we also verify that the decay times obtained from the Randomize Tail- corrected Cone-tracing (RTC) method are more RESULTS ANALYSES accurate than the results obtained from classical methods (Sabine e Eyring) whose values are always Making a general appreciation of all the measured lower than the real ones. acoustical parameters, a first conclusion is the The results of previsional model for speech coincident behaviour of the rooms, and a relative intelligibility evaluation (RASTI Index) are very close independence from the lecture rooms volume. to the measured ones. One of the most significant results obtained from the In order to determine one possible acoustical in situ measurements, are the extreme high values for rehabilitation, we tried in some of the lecture rooms to the reverberation times, specially for the 125 and 250 establish a corrective action using the acoustic octave bands (2.8 - 4.5 seconds). For the 500 to 4000 prediction software. For this proposal we considered a octave bands the reverberation values vary between high performance acoustical paneling for ceilings in 1.9 s (500 Hz) and 1.0 s (4000 Hz). These values are the rear half area of the ceiling (speaker most distant consequence of the coating which has low sound surface), and a 13 mm thick plywood resonant plate absorption coefficients. We obtained the higher values absorber with a regular 5 mm perforation (5,3% of for reverberation in rooms S23 and S25. The ratio total area), mounted over 60 mm air space filled with Volume/Nº of places can be one possible reason. porous material in the back wall of the rooms. Table 2 indicates the limits (maximum and minimum) corresponding to the extreme octave bands (125Hz e 4kHz) for the remaining acoustic parameters CONCLUSIONS measurements: C80 (Clarity – Early-to-late Sound Index), D50 (Definition – Early Energy Fraction), Ts Apparently, by using the simplified prediction (Centre Time) and G (Total Sound Level). model, and for all analysed lecture rooms, we can achieve a reasonable fit with the experimental results, Table 2. Measured Room-acoustical parameter limits not only in the upper octaves 500 Hz, 1, 2 and 4 kHz Room-acoustical Lecture Rooms bands, but also in the lower octave bands 125 and 250 parameter S53, 4, 7, 9, 23, 25 S56, 6, 8 Clarity [-5.0; 4.1] [-6.2; 2.6] Hz. One probable reason for same different results Definition [0.14; 0.58] [0.12; 0.52] may be the inaccurate acoustical absorption/diffusion Centre Time [324; 68] [274; 74] coefficients of the surface coatings and seats, because Total Sound Level [22.5; 29.4] [24.5; 31.1] no previous measurements of absorption coefficients had been carried out. By the measurements of RASTI Index, the mean All the analyzed lecture rooms had similar sound values for each room lye between 0.49 and 0.53, quality parameters. It's possible to improve the corresponding to a FAIR subjective intelligibility. As acoustical behavior fixing absorbent materials only to expected, the front seats have better speech the back wall and the rear half area of the ceiling. transmission results with respect to intelligibility. The reverberation is the main source of interference After the definition and introduction of room in the quality of speech transmission for the analyzed geometry and characteristics on the acoustic prediction lecture rooms. software, we obtain estimated values for the same conditions of the acoustical measurements. Making a general evaluation of the results from the prediction REFERENCES method, we get the best approach in the octave bands 500, 1000 e 2000Hz. The worst results are obtained at 1. Silva, Pedro Martins, Acústica de edifícios, Informação the 125 Hz octave band. This results confirms the note técnica edifícios 8, LNEC, Lisboa 1978. about the expected good resultes in the upper octave bands frequencies. 2. B J Smith, R J Peters and S Owen, Acoustics and Noise For the acoustic parameters the most significant Control, Addison Wesley Longman Limited, England, deviations occurs for the Clarity (C80) and for Early 1996 Decay Time (EDT). We reach the best results for the 3. CATT-Acoustics v7, Room Acoustic Prediction and Desktop Auralization, User’s Manual. Auditorium di Roma - A modern centre in a country with a huge cultural heritage G. Müllera, H. A. Müllera and J. Reinholda aMüller-BBM GmbH, Robert-Koch-Str. 11, 82152 Planegg, Germany

The city of Rome is building a new Cultural Centre. The concept of the centre as well as the acoustical viewpoints deducted from exist- ing, well reputed halls as well as from scientific insights are described. Some acoustical criteria which were referred to during the design of the halls are listed and the general planning procedure including optical and acoustical tests at scale models is shown. Exemplarily key elements of the building and room acoustical design, that are necessary for a high standard, are discussed. The city of Rome is building a Cultural Centre for mu- and a multi-purpose hall for 500 listeners. During the realiza- sic. This Centre consists of a large symphony hall for tion of the project the concept of the small hall was shifted to- 2700 spectators, a mid-size hall for 1200 listeners, which wards an even broader spectrum of uses from concert via op- can be used for chamber music as well as for symphonic era and speech theatre to cinema. music, a multipurpose hall offering a wide variety of uses Taking into account the site on which the halls should be including small operas for 700 spectators and an open-air erected, Renzo Piano's project, which was finally awarded in amphitheatre for an audience of up to 3000 persons. 1994, has as a concept three individual halls positioned around Within his prize -winning design, Arch. Renzo Piano an amphitheatre, which is orientated towards the most silent paid special attention to acoustical viewpoints. These area of the site (Fig. 1). The arrangement of the volumes pro- viewpoints derived from existing, well reputed halls as vides good logistical and acoustical prerequisites for a simu l- well as from scientific insights, constraints given by the taneous use of the halls. The connection of the halls by an an- site and musical presets. They affect seat arrangement, nular connecting building permits the common use of re- positioning of the audience, volume, and general shape of hearsal rooms, technical areas, delivery zones so that both ad- the halls. The large site of construction outside the his- vantages of “all in one building” and “all in separate build- toric centre of Rome made it possible to choose an ade- ings” can be counted on. quate general arrangement of the halls, which permits a highly unrestrained simultaneous use and good acoustical prerequisites. The difficulty of the location was the high noise-impact by the surrounding traffic and the maximum height of the building given by the location near to one of Rome’s hills . HISTORICAL BACKGROUND In Italy, a huge heritage concerning opera houses, mid- size halls and churches offers a large variety for music performances. However, the interest in halls with a large FIGURE 1. General Arrangement of the halls number of listeners taking into account modern aspects of acclimatization, acoustics, comfort, administration and BUILDING ACOUSTICAL ASPECTS organiza tion of performances lead to a growing number From the building acoustical point of view, the separation of of new halls as e.g. in Torino and in Rome. The fact that the three halls will make it possible to use them simu ltane- in the last decade in Rome also existing large halls such ously without any acoustical interferences. The highly quali- as the Auditorio di Via della Conciliazione were trans- fied rehearsal rooms are positioned in between the large halls, formed into large and well accepted concert halls proves sufficiently far from them and generally with a noise protec- that there is a high interest in modern cultural infra- tion system inside. The halls' protection is twofold, against struture. outside and inside noise. Measurements resulted in outside BASIS OF THE PLANNING noise at the site of up to 90-100 dB(A) (peak values of singu- lar events) caused by the heavy traffic on surrounding roads. The original concept of the planning was to build three Thus the delimiting surfaces of the halls have to provide a different concert halls, a large symphony hall for 2700 lis- sound insulation of about 70-80 dB. Along the walls this cre- teners, a midsize hall for 1200 listeners which can be used ated no problems as foyers or acoustically unimportant zones for chamber music as well as for symphonic music generally surround the halls. The roof structure however In order to approach the parameters and to determine the is an acoustical challenge. Due to the limitation of weight general layout of delimiting surfaces, in a first step optical a triple layer structure was chosen (Fig. 2). measurements were carried out. They are aimed to provide a relatively uniform distribution of energy reflected from the ceiling towards the audience. In a second step, room acoustical model tests at the basis of a 1:20 scale model were carried out for the large and the midsize hall.

FIGURE 2. Schematic sketch of the roof structure This structure has varying curvatures and is thus partly extremely sloped. The functioning of the chosen structure and especially the type of fixing between the lower layers without creating bridges was extensively tested by the building company. FIGURE 3. Scale model of the large hall Fig. 4 shows the large hall. Due to its size the sala 2700 is The acoustical protection of the auditoria inside the designed with an arrangement of the audience around the or- building structure is provided as far as possible by a sepa- chestra, a series of reflecting vine-yard-like surfaces reducing rate carrying structure of the halls. By this the number the “effective width” for lateral reflections and two-direction- and area of joints between the "sensitive" and "less sen- ally curved diffuse reflecting ceiling elements providing a good sitive" zones could be minimized. energy distribution sufficiently diffuse in order to reduce the ROOM ACOUSTICAL LAYOUT ris k of delayed strong reflection. OF THE HALLS The planning goal for the room acoustical layout of the hall is aimed to fulfil objective room acoustical re- quirements, which are highly correlated with the subjec- tive perception of musicians and audience. For both groups a background noise level of less than 25 dB(A) is a key prerequisite. The further room acoustical require- ments are described by the known criteria, which are based on (partly) binaural impulse responses: The acous- tical contact among the musicians and the adequate loud- FIGURE 4. Sala 2700 ness in the audience is described by the strength index, Fig. 5 shows the sala 1200. It refers to classical shoe-box- which considerably depends on the size of the halls. The shaped halls, with a stalls area delimited by parallel, diffuse balance between the instruments for listeners and players walls having an optimum distance for lateral reflections. Due correlates with the variance of the strength for different to the reduced distances in this hall the ceiling could be de- positions on the podium. The timbre of the music is de- signed with less diffusion than in the large hall. scribed by the spectral reverberation time, as well as by the frequency dependency of the clarity. The “being sur- rounded” by music is physically described by the value of the interaural cross-correlation coefficient and the clar- ity. The precision of performance and the authenticity of the music for the listeners requires an absence of echoes, FIGURE 5. Sala 1200 a challenge especially considering the size of the large hall. ACKNOWLEDGEMENT Like in all auditoria the acoustical conditions will de- We thank Renzo Piano and his design team for the excellent pend to a certain extent on the location of the seat, so that cooperation and the motivation to tackle acoustical issues con- the individually varying preferences of the listeners can sequently. be taken into account.

Adaptations to a Church for Chamber Music

M.P.M. Luykx M.Sc.a, R.A. Metkemeijer M.Sc.b

a Adviesbureau Peutz & Associes BV, Postbox 66, 6585 ZH Mook, the Netherlands b Adviesbureau Peutz & Associes BV, Postbox 696,2700 AR Zoetermeer, the Netherlands

In the year 1565 the church The Nieuwe Kerk in The Hague, The Netherlands, has been built, nowadays being used mostly as a multifunctional space for art-exhibitions, (dinner-) parties and also concerts. It was desired to adapt this large and reverberant volume of 13.500 m3 and 3,5 seconds in a small concert hall, acoustically suitable for chamber music. Requirements were to shorten the early decay time and increase clarity and intimacy, visually as well as acoustically, as much as possible, but to keep the appearance of the larger volume and late reverberant tail. Therefore it was decided to choose for a semi-open volume build in glass. The adaptations and their acoustical effect and required dimensions have been studied using a computer-model simulation, in which the amount of openings in the smaller, glazed box has been varied. Based on its results, an 60% to 70% closed smaller inner box of 3500 m3 has been designed and implemented, formed by glazed and diffusive elements. Experimental data and listening experience since then show the successful effect. It will give anyone a remarkable experience of being acoustically in a small concert hall but to be visually in a large church, an effect that is opposite of what is usually done by electronic reverberation systems.

INTRODUCTION sufficiently strong and well defined sound. It was clear that in order to reach an increase of early The “Nieuwe Kerk” in the Hague, the Netherlands, sound energy compared to late sound energy and to being build in 1565 was a typical example of a church realize “smaller” acoustics, the acoustical volume had with cathedral-like acoustics. Merely suitable for to be reduced by creating a smaller ‘hall’ inside the organ and Gregorian chant it was nowadays being used church. In that way it was comparable with the earlier mostly as a multifunctional space for exhibitions, project of the fully glazed AGA-hall that was build dinners and also concerts. Planning a total inside the larger volume of the Beurs van Berlage in refurbishment of the church in 1997 there was a Amsterdam, also a cooperation between Zaanen demand to improve its facilities for audience and Spanjers Architects and Peutz. But is this case of the musicians, and especially to improve its acoustics for Nieuwe Kerk, it was required to maintain the late chamber music. The principals “Nederlands reverberance of the church and that the provisions Congresgebouw” and Dr. Anton Philipshall asked the should be removable for organ concerts etc. In order to architects Zaanen Spanjers Architects and Peutz & do so the smaller ‘hall’ clearly could not be fully Associés as acoustical consultants to make suitable closed and should be designed with partial openings. proposals. To investigate acoustically the optimal amount of reflectors and openings in order to gain maximum increase of clarity a computer model was used.

ACOUSTICAL MODELLING

The existing church that has an acoustical volume of about 13500 m3 and a reverberation time of 3.5 s was acoustically not very suitable for chamber music, especially with a limited audience of 450 people. The acoustics were generally judged as muddy, remote and unclear but with a nice sound. In order to make the church more suitable for chamber music it was necessary to bend the indistinctness of the acoustics into more clearness and also into somewhat ‘smaller’ FIGURE 1. Computer model with some reflectors. acoustics, in which smaller ensembles should give a After matching the existing acoustics of the church it was studied how ceiling-reflectors and side-reflectors Energy-time curve, 1 kHz. should be added to create the required volume of 3000 C80= -5 dB. to 4000 m3 and at the same time create more intimacy and clearness. The model, represented in figure 1, was mainly used to investigate to what extent the smaller volume had to be acoustically separated from the main church volume in order to create the acoustic properties aimed for. In this computer simulation study the amount of openings between the small and the main volume was varied between 0 and 100%. It was Integrated energy-time curve, 1 kHz found that optimal acoustics were to be expected if 60 to 70% of the surface area of the smaller volume was closed, assuming that the reverberation of the main volume would be reduced simultaneously to 2 á 2.5 s. Calculated impulse-responses showed significant increases of early reflected sound energy, leading to a expected increase of loudness of 2 dB and an expected increase of 3 to 4 dB of the clarity-index (C-80, measure for clarity and intelligibility of music). FIGURE 3. Impulse-response measurement before. Based on these results glazed and diffusive elements have been implemented above and around the seating area, see figure 2, and additionally retractable curtains Energy-time curve, 1 kHz have been applied. C80=+2 dB

Integrated energy-time curve.

FIGURE 2. Plan of church with added reflectors.

The curved, glazed elements have been partially FIGURE 4. Impulse-response measurement after. intersected with wooden panels to obtain additional sound diffusion and to increase visual intimacy. The increase of clarity is confirmed by listening experience, in which also the late reverberance of the RESULTS AFTER REALISATION larger church volume can be experienced at for instance music stop chords. It will give anyone a Based on the acoustic parameters and reflection remarkable experience of being acoustically in a small patterns measured an acoustical smaller volume of concert hall but to be visually in a large church, an about 3500 m3 can be deduced, which meets the effect that is opposite of what is usually done by requirements to realize the acoustics of a small hall. electronic reverberation systems. Comparison of impulse-responses measured before The resulting acoustics after the refurbishment have and after measures taken, shows that an increase of the been widely appreciated, and the musical use of the clarity-index (C-80) of ca. 4 to 5 dB has been reached. hall has been growing ever since. Even recording An example of two impulse responses before and after companies seem to have discovered its precise and is given in figures 3 and 4, in which the addition of clear though living acoustics. early reflections can be seen.

A Design Principle for Stage Acoustics in Concert Halls Z. Maekawaa , Y. Kawaib, and Y. Harac

a Environmental Acoustics Lab. Faculty of Eng. Kobe Univ., Rokko Kobe 657-8501 Japan b Dept. of Architecture, Faculty of Eng. Kansai Univ., 3-3-35 Yamate, Suita 564-0073 Japan c OTO Acoustic Eng. Assoc., Nissei-Buil., 1-1-3 Kitahorie, Nisiku, Osaka 550-0014 Japan

By experimental studies with a scale model of an end stage, it is shown that the stage acoustics can be controlled by the style and slits on the wall surface of stage enclosure in a typical shoe box type concert hall especially in low frequency range. Also, by theoretical studies it becomes clear that the values calculated by BEM are useful for the same purpose.

INTRODUCTION 1100mm 140

Among recording engineers and musicians there are 200 latent complaints about stage acoustics at newly 530 660 660 constructed concert halls in Japan, such as “the 530 acoustics is unnatural”, “there is some lack of 200 Slit 926 definition and clearness”, or “there seems to be a Elevation of Back wall Section choked or stifled tone in feeling”, etc.. Almost all of these concert halls have an end stage ƒÆ 167 S1 167 enclosed tightly with reflecting walls and ceiling, in Slit order not to lose the sound energy. It is considered that the complaints are brought about by the adverse effects 926 Unevenness of acoustic reflections from boundaries, such as a kind P2 133 of booming or coloration caused by interference. 33 1200 Therefore, the control of acoustic reflections is very Plan essential to give a good stage acoustics in a concert Figure 1. Drawings of the stage model hall, especially in low frequency range. [1] Among room acousticians, it is said that the tone Table 1. Configuration of scale models quality of a concert hall is depends almost entirely on Side Walls Back Wall the early reflection until about 100ms after the direct Case θ uneven Slits Slit sound arrives. If so, the acoustics on a concert stage F 0 0 A 0 30mm will be under the control of the reflections only from B 0 30 30mm stage enclosure itself, since reflection from the C 0 30 30 30mm audience space must be arriving later. Therefore, in D 7 30 order to study the tone quality of stage acoustics, it F 7 30 30 30 should be possible to use the stage enclosure only, without audience space. (B&K type4135). The early stage of IR was trans- formed in frequency domain and shown with the SCALE MODEL EXPERIMENTS curves between 20-80 Hz, converting to the real size frequencies. Representative results are shown in Fig.2. A 1/15 scale model of a stage only ,in a typical shoe An evaluation can be performed comparing box type concert hall with 800 seats, was used for unevenness of curves between 40-80 Hz, the smoother acoustic measurements with various wall styles the curve the better. containing uneven surface and slits, as shown in Fig. 1. Comparing the results of cases A & B, the slits on 1 and Table 1. the side wall are effective to make the dip smaller Acoustic measurements were performed to at near 60 Hz, as shown in Fig. 2-a. obtain the impulse response(IR) with a system of 2. Comparing the results of cases B & C, the slits on the MLSSA (DRA Lab.), using a 50mm the back wall makes the deviation of curve electrodynamic loud- speaker (BOSE MM-2), and a smaller, as shown in Fig. 2-b. condenser-microphon   Case A BEM
 

  Case B

Model



 

Fig. 2-a  Interval: 0.3Hz



 

Case B 
     

  Case C
 Interval: 1.2Hz  Fig. 2-b Fig. 3-a Case C 
     

Case A




   


  

Case C  BEM 
    

Fig. 2-c
 

Model

Case A

  

 Interval: 0.3Hz

 
 

Case D 

    

Fig. 2-d 



 Case C Interval: 1.2Hz 

 Fig. 3-b Case F
      Case E



  Fig. 2-e
 Figure 3. Measured transfer characteristics.        And calculated values by BEM


  Figure 2. Measured transfer characteristics In the stage models. BEM method. This method shows the results in the steady state. However, it may be interesting to what extent they agree with the results of model experiments shown above. Figure 3 shows the representative results 3. All slits are clearly effective to make the curve comparing the experiments. It can be said that both smoother, as shown in Fig. 2-c. results almost agree in style, though there are some 4. Comparing the results of cases of A & D, the effect differences in detail. Theoretical results, therefore, of the angleθof side wall is very useful for the might be possible as an alternative in place of model same purpose as shown in Fig. 2-d. experiments. 5. Comparing the results of cases of E & C, the effect of the angleθis not so big, because of the effect of ACKNOWLEDGEMENT slits on all walls, as shown in Fig. 2-e. 6. Comparing the results of cases of F & A, the Model experiments were performed in an results did not show such a difference. But the anechoic room in Kyoto University with the unevenness on the wall surface was adopted for kind help of the staff. the sake of higher frequencies. REFERENCES THEORETICAL CONSIDERATION [1] Y. Hirasawa and Z. Maekawa; An evaluation Theoretical calculation has been done on the same method of concert hall acoustics. 16thICA(1998) 2125 conditions shown in Figure 1 and Table 1 by using The Acoustic Conditions in Finnish Concert Spaces

H. Moller, T. Lahti and A. Ruusuvuori

Akukon Oy Consulting Engineers, Kornetintie 4, 00380 Helsinki, Finland

The paper will describe a series of acoustic measurements currently undertaken in Finnish concert halls. The measurement result will be used in a comparison of the acoustic conditions in Finnish concert spaces with other concert spaces.

INTRODUCTION halls has previously been documented, see [1], [2] and [3], but the intention of the project is to make a unified It seems to be a common belief in Finland that "we database for the halls. have the worlds best musicians but the worlds worst concert halls". This indeed very strange as it is hardly Center name Seats possible to find any other country which has the same Tapiola Hall, Espoon Kultuuriksekus, Espoo 780 Helsingin Konservatorio, Helsinki 500 amount of concert seats per capita. The major part of Järvenpää talo, Järvenpää 570 these halls has been built within the last 20 years; a Kaukametsän kultuuritalo, Kajaani 550 total of about 27 halls of 350 seats or more. This is in Kansantaidekeskus, Kaustinen 390 a country with about 5.2 million people. And there is Kuopion Musiikkikeskus, Kuopio 1060 Laurentius sali, Lohja 473 still new halls being constructed. The Sibelius house in Mikaeli kultuurikeskus, Mikkeli 690 Lahti opened in the spring of year 2000 and currently Oulun musiikkikeskus, Oulu 816 there are at least 5 new halls in design as well as the Promenaadikeskus, Pori 700 preliminary design for the new Helsinki Music house, Large Hall, Tamperetalo, Tampere 2000 which will be the third 1000+ seat concert hall in Chamber Hall, Tampertalo, Tampere 500 Sigyn-sali, Turun Konservatorio, Turku 350 Helsinki, a city with approximately 1 million Martinus sali, Vantaa 410 inhabitants. Hyvinkään Kultuurikeskus, Hyvinkää 450 Iisalmen Kultuuritalo, Iilsalmi 470 The size of the halls of this survey varies in size from Imatra Kultuurikeskus, Imatra 530 2000 to a typical about 400. In some cases, the halls Carelia sali, Joensuun Yliopisto, Joensuu 600 Kuusamo talo, Kuusamo 530 represent the only performance space in the Kuusankoski talo, Kuusankoski 510 community, in many other larger cities there are at Kuhmon talo, Kuhmo 670 least one older theater as well as the multi-purpose Lappeenrantan talo, Lappenranta 690 hall. Lieksän kultuuri keskus, Lieksä 400 Poleeni, Pieksämäki 350 Kauppaporvarin Kulttuuri- ja kongressikeskus, Raahe 420 From an international point of view, it is of course Rauma-sali, Rauma 435 questionable to call a 350-seat space a concert hall. We Seinäjoki-sali, Campus-talo, Seinäjoki 360 do however take the liberty to make such a Table 1: The cultural with 350+ seat halls build in Finland classification as all the halls, even though "marketed" between 1980 and 1999. Measured halls are in italic. as multi purpose halls, has been build with symphonic music and orchestras as the primary design goal. This means that the stage size has been determined by the size needed for a symphony orchestra and especially THE MEASURED HALLS that pure acoustic design and details has been given higher priority than the needs of the sound-, light and So far 14 halls has been measured, see Table 1. All A/V systems. The volume of all the halls has been halls has been empty during measurements. The stage dimensioned to give appropriate reverberation time for has either been empty or the chairs has been moved so acoustic music. that they obstruct sound paths as little as possible.

The measurements described in this paper is a part of a The measured halls represent quite typical gross- project to is to make a database with acoustic data for shapes: most are more or less shoebox shape, in some all Finnish halls with more than 350 seat. Some of the cases with slightly angled walls, if not quite fan Reverberation time, s shaped. In most cases the ceiling and the stage is 3 designed in a "directed reflections" manner. This means that the ceiling, especially over the stage is slanted to reflect sound energy to the audience. Most of the halls have back balconies or both back- and 2 side-balconies.

The largest hall has 2000 seats and the smallest 350 seat and the average seat count is 600 seats. 1

RT 20

EDT CONCLUSIONS AND FUTURE WORK 0 125 250 500 1000 2000 4000 f, Hz Even though the measurement series is not yet Figure 1: Measured reverberation in the Tamperetalo, Large hall complete, it is fair to say that quite a few halls with Reverberation time, s "good acoustics" has been found. When compared with 3 data from equivalent size halls, the Finnish halls actually looks quite good. The main problem from the larger orchestras point of view is that most of these halls are relatively small. Even though the data for the 2 tree "old halls" are not shown in this paper, it is fair to conclude that none of these nor the large hall at Tamperetalo, has room acoustics to match the best 1 halls. RT20 RT20 SOFT 1 Another conclusion is that in general the changeable RT20 SOFT 2 acoustic system does not work or at least does not EDT 0 produce any significant change in the acoustic 125 250 500 1000 2000 4000 f, Hz conditions in the hall. The only real exception, is the Kaukametsä hall in Kajaani in northern Finland, where Figure 2: Measured reverberation in the Kaukametsä Hall, a hall with changeable acoustics the use of all the changeable absorption in the audience area and on the stage, gives a significant change in the REFERENCES acoustic conditions, see figure 2. 1. Lahti T & Möller H, "Practical experience with concert hall designs with a computer, featuring the glass- The next steps will be to complete the measurement walled Sigyn Hall of Turku". Auditorium Design at the and analysis of the halls. Also we intend to make Millennium, Belfast 22-24.5.1997. Proceedings Institute subjective testing in some of the most interesting halls. of Acoustics Volume 19 Pt3 (1997), 11–18. 2. Möller H, Lahti T & Ruusuvuori A, "New design of medium-sized concert halls", Auditoria: The legacy of the 20th Century and beyond 2000, Manchester 22- ACKNOWLEDGMENTS 24.10.1999, Proceedings Institute of Acoustics Volume 21 Pt 6 (1999), 117-122 We would like to thank the people at the Helsinki 3. Beranek L, "Music, Acoustics and Architecture", University of Technology working on the New York, 1962, John Wiley and Sons VÄRE/TAKU project as well as our partners in the VÄRE/BINA project for their cooperation. Also we 4. Peltonen T, Lokki T, Gouatarbes B, Merimaa J & would like to thank the technical staff of the cultural Karjalainen M. "A System for Multichannel and Binaural houses for their help and support Room Response measurements", AES 110th Convention, Amsterdam, The Netherlands 12-15.5.2001, Preprint 5289.

5. Gade AC, "Akustik i danske koncertsale", Publikation nr. 22, Laboratoriet for Akustik, DTH, Lyngby 1984.

Relation between Image-Split and Listener Envelopment

M. Morimotoa, K. Nakagawaa,b, M. Jinyaa and M. Kawamotoa aEnvironmental Acoustics Laboratory, Faculty of Engineering, Kobe University, Nada 657-8501 Kobe, Japan bEngineer of Environment Division, Nikken Sekkei Co. Ltd., 541-8528 Osaka, Japan

The authors have suggested a hypothesis that the components of reflections beyond the upper limit of the law of the first wave front contribute to listener envelopment (LEV). In this paper, listening tests were performed to examine the hypothesis for its validity. Test sound fields consisted of a direct sound and two reverberation signals. In the listening tests the relative sound pressure level of reverberation signals to the direct sound was changed randomly. Firstly, the thresholds of image-split, which corresponds to the upper limit, and LEV were measured. The results did not demonstrate a significant difference between those thresholds, which support the hypothesis. Secondly, the thresholds of echo and echo disturbance were measured. The results showed that the threshold of echo disturbance was higher than that of LEV perception by about 20dB. This suggests a possibility to create rich LEV without echo disturbance.

image-split and LEV, echo disturbance and echo INTRODUCTION perception, respectively.

The task of the subject was to map all sound images Several papers have demonstrated the relations which he perceived in case of the threshold of between listener envelopment (LEV) and temporal image-split and to answer whether he could perceive characteristics of sound fields[1-5]. However, they are each auditory phenomenon in the other thresholds, not necessarily in agreement. It is important to make after each presentation of stimulus. Each subject was clear necessary conditions for the perception of LEV tested 51 times for each stimulus. The listening test of from the standpoint of auditory event in order to four kinds of threshold was performed separately in the explain the contradiction. order image-split, LEV, echo perception and echo Morimoto and Iida [6] show that the acoustic disturbance. Four male students acted as subjects for all components under the upper limit of the law of the first tests. wave front contribute to auditory source width which is one of two components of spatial impression as well as LEV. From this result, a hypothesis that the DATA REDUCTION components of reflections beyond the upper limit of the law contribute to LEV can be suggested [1,4]. The data reduction were done separately for each In this paper, to examine the hypothesis for its subject. All thresholds was obtained by using the validity, four thresholds were measured by the listening normal-interpolation process. The process is explained tests: image-split which corresponds to the upper limit, showing an example of the threshold of image-split for LEV, echo perception, and echo disturbance. subject B in Fig. 1. The percentage of image-split was

obtained for each stimulus. Z-transformation of those

METHOD percentage were performed and the regression line and the correlation coefficient were obtained, neglecting

data less than 1.0% and more than 99.0%. The The motif used for the listening tests was a 7 s threshold (value at z=0) and its standard deviation section of the 1st movement of Mozart's Divertimento (values at z=±1) were obtained by regarding the data as recorded in an anechoic chamber. The sound field used being normal distribution because the correlation as a stimulus consisted of a direct sound placed in front coefficient was almost 1.0. The coefficients for all and two reverberation signals placed at ±135o . Their thresholds and for all subjects exceeded 0.928. reverberation times were constant at 2.0 s and their frequency characteristics were flat. Reverberation delays were 80 and 81ms. The sound pressure level of RESULTS AND DISCUSSION the direct sound were kept constant at 70dBA and the relative sound pressure level (∆L) of the first Figure 2 shows measured values of four kinds of component of reverberation signals to the direct sound threshold with their standard deviations together for were changed in random order. ∆L were set at 11 steps each subject. There is little difference between from -39.6 to -19.6dB, at 9 steps from -11.6 to -3.6dB individuals for all four thresholds. The difference and at 11 steps from -52.6 to -42.6dB for thresholds of

upper limit. In other words, it is necessary to provide reflections beyond the upper limit in order to make a listener perceive LEV. Meanwhile, the threshold of echo disturbance is higher than that of LEV by about 20dB. This means that reflections beyond the threshold of echo do not always occur disturbance, but contribute to the perception of LEV.

Z - Value Z - Value CONCLUSION

The results of listening tests support the hypothesis that the components of reflections beyond the upper limit of the law of the first wave front contribute to LEV. Relative SPL of Rev. to a direct sound, ∆L (dB) FIGURE 1. method of data reduction. An example of REFERENCES subject B for image-split. [1] J. S. Bradley and G. A. Soulodre, J. Acoust. Soc. between thresholds of image-split and echo perception Am. 97, 2263-2271 (1995). is about 20dB. This means that the subjects could [2] J. S. Bradley and G. A. Soulodre, J. Acoust. Soc. discriminate between them. Am. 98, 2590-2597 (1995). The difference between image-split and LEV is [3] J. S. Bradley, R. D. Reich and S. G. Norcross,J. small for any subject. The threshold of LEV is within Acoust. Soc. Am. 108, 651-661 (2000). the standard deviation of image-split except for subject [4] M. Morimoto, K. Iida and K. Sakagami, Applied B. From these results, the threshold of image-split and Acoustics 62, 109-124 (2001). LEV can be considered to be equal. This supports the [5] T. Hanyu and S. Kimura, Applied Acoustics 62, hypothesis that the components of reflections beyond 155-184 (2001). the upper limit of the law of the first wave front [6] M. Morimoto and K. Iida, Proceedings of Institute contribute to LEV, since image-split corresponds to the of Acoustics 14, 85-91 (1992).

) dB ( L ∆

,

Relative SPL of Rev. to a direct sound a direct to Rev. of SPL Relative Subject FIGURE 2. Four kinds of threshold and their standard deviations. Open circle; image-split, closed circle; listener envelopment, open triangle; echo-disturbance, and closed triangle; echo.

Acoustical, musical and social factors in the resolution of early acoustical problems at the Concertgebouw

Pamela Clements

Kirkegaard Associates, 801 West Adams Street, Chicago, IL 60607 USA

When the Concertgebouw opened in 1888 it was found to have significant acoustical problems. Balance between brass and strings was poor, and the room was considered overly “resonant”. Poor attendance – the result of one inadequate access road - exacerbated the problem. The missing organ was blamed, but when the organ was installed in 1891 many thought the sound even worse. Resolution came progressively, and involved a serendipitous combination of renovation of the stage, improvement in the orchestra’s performance standards, the broad interpretive conducting style of Willem Mengelberg, the popular appeal of Tchaikovsky’s 6th symphony, and new roads which brought enthusiastic audiences to fill the hall.

INTRODUCTION and to make the ceiling reasonably high, which it was felt was necessary for acoustical quality. The hall certainly is reverberant, with a measured RT Conceived as a grand concert facility that would foster the musical life of Amsterdam, the Concertgebouw at mid frequencies of 2.0 seconds occupied and around was designed with close attention to acoustical quality. 2.58 seconds unoccupied, with a very strong bass response.[3] These qualities in the hall are now very The large hall was large indeed, with seats for an audience of over 2000, and a high stage with steep much admired throughout the musical world, but amphitheater-style risers designed for an orchestra of audiences and orchestras in Amsterdam of the late 1880s were unaccustomed to this sound environment. 120 and chorus of 500. The only acoustically excellent hall in Amsterdam After the two major openings – that of the large hall on 11 April 1888 and the first concert of the new prior to the opening of the Concertgebouw was the concert room at the Felix Meritus Society, which was a Concertgebouw Orchestra on 3 November 1888 - it small, oval room for chamber music – a very different was clear that the hall was less than perfect acoustically. Many considered it too “resonant” and space acoustically. The other predecessor, the Parkzaal, which with 2000 subscribers influenced the size of the though the string sound was beautiful it was dominated Concertgebouw, had terrible acoustics. by the brass instruments.[1] Over the next twelve years the acoustical problems were gradually resolved, The unfamiliar “resonance” in the Concertgebouw was exacerbated as a result of an apparently unrelated through a fortuitous combination of successful issue of urban development The Concertgebouw was renovation, musical events and social change. Fortunately a significant amount of evidence has built on land in the municipality of Niewer Amstel, just outside the city limits of Amsterdam proper, in the area survived on the early years of the Concertgebouw and that developed into the museum and arts sector of the design of the hall by its architect A.L. van Gendt. This paper is based on evidence in scholarly historical Amsterdam. The building of access roads and horse tram routes was delayed by wrangling between the works including those published under the auspices of municipalities, and when the hall opened there was the Concertgebouw, research by architectural historian Lydia Lansink, and on documents held in the only one access road. It took so long for the carriages to leave after the opening concert that many people Amsterdam Municipal Archives.[2] cancelled their subscriptions, and thereafter for many

years the concerts were plagued with very poor EARLY REMEDIES attendance. Reverberation time was thus longer, and was likely made worse by the large areas of flat floor The “resonance” of the Concertgebouw – its long left exposed when management removed the reverberation time – is largely a function of the size of unoccupied chairs to help reduce the sense of the hall and its geometry. This resulted from early emptiness. decisions of the founding committee: to build the hall In the early years the orchestra struggled with the for an audience of 2000, to make the room reasonably room’s acoustical character. Not only was the wide so that it would not become excessively long, to reverberation unaccustomed, but the steepness of the size the stage for the very large orchestra and chorus, risers made balance between brass and strings almost impossible to achieve. The orchestra was newly approach allowed a flexibility of dynamics and tempo formed and not yet of the highest caliber, and with 66 that could exploit the hall’s acoustical character. The members it was still relatively small. Moreover, the late Romantic repertoire performed in a broadly choral seating at either side of the stage was empty. interpretive style was well suited to the hall’s Initially it was thought that perhaps the difficulties reverberant character, and the public was now with balance were being caused by the missing organ, enraptured by the musical experience. and the curtain that covered the space where it was to As it turned out, the balance problems had not been be located. But when the organ was installed in 1891 solved with the Pathétique. Mengelberg now urged there was no resolution: some felt that the acoustical what others had suggested – rebuilding the stage. quality had improved but other felt that the balance Though management feared such a major renovation in was worse, and that the sound quality had now become case it turned out to be detrimental rather than “harsh”. [4] beneficial, eventually the work went ahead. In 1899 the Other remedies were tried progressively. Potted stage was lowered drastically by 20 cm at front, 2.3m plants were placed on stage and in the audience area to at the rear, and 1.2 m at the sides, and a wall was “soak up” the sound. Low wooden screens were created behind the orchestra at the base of the organ. installed at the rear of the stage in an attempt to reduce The renovation was a great success: orchestral balance its size. Curtains were installed at the doors and fabric was now easy to achieve. The Concertgebouw’s was draped over the screens. Carpet was installed acoustical problems were solved. under the percussion and brass, and it was desired also to put carpet in the aisles but there were insufficient ACKNOWLEDGMENTS funds for this. With poor attendance and a reputation for “something lacking” in its acoustics, the Special thanks are due to Jan Henk van Weerd and the Concertgebouw in its early years suffered extreme Concertgebouw for assistance with this research and financial hardship. for permission to use photographs in the presentation

slides. Lydia Lansink’s research and writing on the RESOLUTION 1896-1899 architecture of A.L. van Gendt has also been invaluable. In 1895 Willem Mengelberg was appointed to succeed Willem Kes as conductor of the Concertgebouw REFERENCES Orchestra. Mengelberg was only 24 years old, and there was some doubt in the Amsterdam community as 1. Lydia Lansink, “Het Gebouw: functie en vorm,” in to the suitability of his appointment. But these fears Historie en kroniek van het Concertgebouw en het were put to rest with one remarkable concert on 24 Concertgebouworkest 1888-1988, Zutphen: De Walburg September 1896. This was the first Concertgebouw Pers, 1989, vol. 2, p. 90. performance of Tchaikovsky’s Sixth Symphony, the Pathétique. The concert establis hed Mengelberg’s 2. Evidence in this paper and in the presentation is largely ability as a conductor, and proved the orchestra’s drawn from: S.A.M. Bottenheim, Geschiedenis Van Het greatness by demonstrating that it could rise to the Concertgebouw, Amsterdam: Joost Van Den Vondel, demands of modern composition. Mengelberg was 1948; Historie en kroniek van het Concertgebouw en het Concertgebouworkest 1888-1988, Zutphen: De Walburg even able to keep the balance between the brass and Pers, 1989, 2 vols; Lydia Lansink, “De akoestiek van het the strings despite the hall’s acoustics.[5] Concertgebouw historisch bezien,” Preludium, 36(8), 35- The concert marked a turning point in the 45 (April 1978); Otto Glastra van Loon, Onder de stenen Concertgebouw’s history. Audience enthusiasm for lier: Het Concertgebouworkest, Amsterdam: Uitgeverij this “truly inspirational” musical experience was so Ploegsma, 1969; and Wouter Paap, Willem Mengelberg, great that the performance was repeated thirteen times, Amsterdam: Elsevier, 1960, tr. R.H. Hardie, P.M. and thereafter audiences clamored for concert tickets. Schouten and D. Tait. Fortuitously, there had been a change in municipal boundaries in 1895 which resolved the delays in road 3. See data in Leo Beranek, Concert and Opera Halls: How they Sound, Woodbury, NY: Acoustical Society of building. Newly enthusiastic audiences now poured America, 1996, p. 613. down newly built roads to fill the hall. Mengelberg’s broad, free “Wagnerian” style of 4. Otto Glastra van Loon, Onder de stenen lier: Het interpretation had captured the imagination of Concertgebouworkerst, p. 13. Amsterdam’s musical public. The orchestra’s standards, and hence the musicians’ ability to respond 5. S.A.M. Bottenheim, Geschiedenis Van Het to and work with the acoustical conditions, were now Concertgebouw, pp.13-17. rapidly improving, and Mengelberg’s conducting

Redevelopment of the Sydney Conservatorium of Music Part I: Design imperatives

R. L. Kirkegaarda and S. Prettyb

aKirkegaard Associates, Consultants in Architectural Acoustics, 801 Adams Street, Chicago, Illinois 60607, USA bSydney Conservatorium of Music, The University of Sydney, Sydney NSW 2000, Australia

The Sydney Conservatorium of Music occupies a prominent site within a park setting near the Sydney Opera House. Ever since its inception, however, the facility and site have struggled to maintain pace with the Conservatorium’s constantly-expanding reputation, educational mission and enrollment. Redevelopment of the Conservatorium of Music and Conservatorium High School has resulted in fine accommodations for music tuition, rehearsal, and performance while preserving heritage elements and the surrounding landscape.

HISTORY OF THE CONSERVATORIUM torship of Eugene Goossens from 1946 to 1956 was par- ticularly remarkable, but by the end of his tenure, the Established in 1915 as the New South Wales State inadequacies of the stables building were painfully clear. Conservatorium of Music, the Sydney Conservatorium of During the period 1962 to 1984, seven separate en- Music is one of Australia’s oldest and most prestigious quiries into the Conservatorium’s accommodation issue cultural institutions. With its roots firmly in colonialism, were held, each underlining the inadequacies of the exist- the early years saw an institution dominated by European ing premises in terms of lack of space and sound isola- and British influences. Indeed the building allocated for tion between spaces, rising damp, and the intrusion of the Conservatorium was the stables for the former British noise and vibration from the adjacent underground rail governor, and the first Director, appointed in 1915, was tunnels. Accretions unsympathetic to heritage and envi- Henri Verbrugghen, a Belgian-born violinist and conduc- ronmental issues were applied to the stables in well- tor of distinction. intended efforts to improve the situation. In the early 1980’s, sections of the Conservatorium’s activities were removed from the stables site to largely unsuitable office space, and in 1990, the Conservatorium was required to amalgamate with a University. The sub- sequent alliance with The University of Sydney applied a new layer of political and academic complications that, in conjunction with the split Conservatorium campus, further exacerbated the accommodation crisis. After a lengthy and heated review of alternative sites

for the Conservatorium, redevelopment of the original FIGURE 1. Ground level plan (1915). stables site was funded in 1997. All activities of the Con- servatorium located within the stables, including the Verbrugghen attracted important musical figures Conservatorium High School, were removed to the Aus- from abroad to staff the institution and enrolled nearly tralian Technology Park, an adaptive reuse of inner-city 800 students by the beginning of the 1917 academic year. railyard sheds, while the works progressed. His initiatives to reach out to school-age students re- Unfortunately, the extent of heritage items remaining sulted in the 1919 founding of the Conservatorium High from the early colonial settlement of the site had been School, a publicly-funded institution offering secondary- underestimated, and so, as excavation got underway, aged students an opportunity to combine their academic controversy about the suitability of the site for redevel- studies with intensive musical training. opment broke out. The cost of delays and additional During the ensuing years, the Conservatorium bene- work to complete the redevelopment, due largely to heri- fited from the leadership of many prominent national and tage and environmental issues, eventually required more international figures as Director. The period of the Direc- than a doubling of the original allocation of funds. REDEVELOPMENT OF THE teachers and students in a building sandwiched between CONSERVATORIUM railway and expressway tunnels for too much of the day. Alternatives were needed that would allow Conservato- rium occupants to enjoy stimulating environments free of In order for the redevelopment to be a success, the noise and distraction. design team was required to reconcile a broad spectrum The architects undertook extensive replanning so that of design imperatives. The stables were to be reinstated the most sensitive recording spaces could be relocated as a freestanding building. Verbrugghen Hall, the concert away from the railway tunnels, and where high levels of hall that had occupied the stables courtyard since 1915, groundborne vibration were still anticipated, various por- was to be retained as the centre of the new Conservato- tions of the building were decoupled from the underlying rium. Further erosion of public lands was to be avoided. rock with springs or elastomeric pads. Teaching spaces were to represent best international Many teaching studios were moved close to exterior practice in music education standards. All of this was to walls and interior courtyards that served as light and air be achieved within strict heritage constraints and in re- wells. Natural light was brought into the remainder of the sponse to sometime hostile public pressure groups. studios and practice rooms by placing full-height side- With input from other design professionals, including lights alongside all doors at the corridors. architect Phillipe Robert and acoustician R. Lawrence The benevolent Sydney climate encouraged the use of Kirkegaard, Government Architect Chris Johnson lodged fancoil units in conjunction with untempered outside air a Development Application to replace the existing exten- in all Conservatorium spaces other than performance sions with buildings on three sides of the original stables spaces. Three-stage speed selection allowed users to con- after extensive excavation of the underlying sandstone. trol personal comfort and varying degrees of acoustical Topped with landscaped terraces, these buildings con- privacy through increased levels of ambient noise. tained an opera house, recital halls, lecture theatres, Room acoustical advantages were realised by com- teaching studios, practice rooms, the High School class- bining room functions and areas. For example, when the rooms, and a canteen. A library and car park were lo- stagehouse was removed from the opera house and its cated entirely underground beneath a new landscaped floor area was added to that of a modestly-sized rehearsal plaza that connected the complex to the city. A glazed hall, the Music Workshop was born. Sections of retract- atrium connected the heritage building to the new facili- able seating, an orchestra pit cover, canvas ceiling reflec- ties and formed a public foyer. tors, and adjustable absorption systems, combined with a

generous room volume, resulted in a more flexible space that could now support lectures and multi-media produc- tions as well as opera and large ensemble rehearsals. Similarly, the area of a lecture theatre was combined with that of a small ensemble rehearsal room to form a second Recital Hall. Separated by a lounge, the two re- cital spaces could now be efficiently cycled back and forth during examination periods. Verbrugghen Hall, however, could not be relocated or entirely rebuilt because of heritage concerns. Its history and transformation into a concert hall are examined in

Part II: Integrating acoustical and architectural solutions. FIGURE 2. Level –1 plan (2001). ACKNOWLEDGMENTS The Development Application scheme, however, did not resolve all design conflicts. It had been proposed that The State Government of New South Wales provided Verbrugghen Hall be lowered 7.5m to improve isolation funding for the Redevelopment. The Department of Edu- from environmental noise and to match the elevation of cation and Training provided project direction, while its platform with that of the other performance spaces, project management was provided by The Department of but cost, structural and heritage concerns eventually Public Works and Services. forced its return to a ground level position. Furthermore, Wilkinson Murray Pty Ltd served as Building Isola- this conceptual design was thought to bury too many tion Consultants. Redevelopment of the Sydney Conservatorium of Music Part II: Integrating acoustical and architectural solutions

E. McCuea and B. McGregorb

aKirkegaard Associates, Consultants in Architectural Acoustics, 954 Pearl Street, Boulder, Colorado 80302, USA bDaryl Jackson Robin Dyke Architects, 64 Rose Street, Chippendale NSW 2008, Australia

At the heart of the 30,000m2 Redevelopment of the Sydney Conservatorium of Music and Conservatorium High School lies Ver- brugghen Hall. Situated within the former courtyard of a stables building, every aspect of the renovation of Verbrugghen has been guided by the Heritage Act. The result is a concert hall that conserves archaeological treasures beneath an ample performance plat- form for a full symphony orchestra and audience seating for up to 550.

GOVERNMENT STABLES construction was rudimentary, and many elements of earlier buildings on the site were simply concealed under The Government Stables were constructed between a layer of plaster, timber or soil. 1817 and 1820 by Governor Lachlan Macquarie as part In 1956, trains began to run in two City Circle rail- of a building programme within the newly-founded Brit- way tunnels constructed immediately to the east of the ish settlement of Sydney. It was to provide horse stables building, and soon after the Cahill Expressway tunnel and servants’ quarters within a harbourside estate. The was constructed to the west. architect for these buildings was Francis Greenway, a Bristol architect transported to Australia as a convict VERBRUGGHEN HALL following a commuted death sentence for forgery. Government Architect Seymour Wells’ plan for the concert hall was based upon a 31m long and 18.5m wide rectangle terminated by a semi-circular apse. Sound- projecting walls were provided at the sides of the per- formance platform equipped with chorus risers, and an orchestra pit was partially recessed under the stage. An audience of 850 was seated on two levels – the upper being a wedge-shaped balcony with extensive overhang of the stalls. Lead-light windows, a coffered and boarded ceiling 12.5m above the platform, and spare decorative plasterwork completed the scheme. This shoebox-shaped hall was retrofitted in the FIGURE 1. Stables within the landscape (1827). 1960’s with a proscenium and modest theatrical rigging systems in order to accommodate operatic productions. With labour provided by convicts, construction of the In the 1970’s an organ was commissioned to fill the apse. stables commenced in 1817 in an English gothic style At this point Verbrugghen Hall was struggling to around a central courtyard. From 1820 until Australian serve as a theatre, as a concert hall, and as an assembly Federation in 1901, the stables served the British gover- hall for formal academic convocations. Generations of nors. In 1915 it was chosen to house the new Conservato- Sydneysiders were defining their musical experiences in rium of Music. terms of the acoustics of Verbrugghen. Those experi- An 850-seat concert hall was inserted into the central ences, however pleasant, were marred by the shortcom- courtyard while teaching and support facilities were lo- ings of the hall’s hybrid design. Verbrugghen Hall was cated in the refurbished stables and quarters. Because this falling well short of the world-class reputation of the adaptation was carried out during the stringent war years, Sydney Conservatorium of Music. Design Integration its divisions reorganised in order to fit within the reduced height of the upstage niche. The Australian system of heritage assessment deter- Subtle reshaping of the sidewalls and the careful mined that Verbrugghen Hall was of “high cultural sig- specification of the audience seating avoided colouration nificance”. The goals of the Conservatorium redevelop- and uneven decay of sound at the seating plane. Fine- ment therefore compelled the design team to extract from tuning of the reflection sequences in the upper volume the historic fabric a new concert hall that could be fa- was made possible with the installation of adjustable vourably compared to the halls that serve the world’s absorption systems of vertically-tracking banners and most highly-regarded schools of music. horizontal draperies.

FIGURE 3. First floor plan (2001).

FIGURE 2. Longitudinal section (2001). After demolition of the proscenium and the raked stage floor, the resulting 18.5m room width made possi- To achieve reverberation and energy density appro- ble the design of a 13.7m deep platform and riser system priate to various performance types, additional acoustic that could accommodate a full symphony orchestra. In- volume and solidity of construction were required in tended to enable each musician to see and hear every spite of the fact that the existing walls were in places other player, the risers eliminated the acoustic necessity 900mm thick. Replacement of much of the lightweight of an overhead canopy of reflectors. In this way, both timber ceiling with a sound-transparent screen material good ensemble conditions and unobstructed views of the allowed the attic volume to be fully accessed and enabled historical ceiling plane and organ niche were realised. a displacement air distribution system to be pursued. With the sound-robbing boarded ceiling out of the way, precast concrete roof panels on new trusses could enhance the bass response in the hall and help isolate the interior from the distraction of indigenous birdcalls originating in the adjacent Royal Botanic Gardens. Dou- ble-glazed windows, however, were also required to fully mute the raucous cockatoos, kookaburras and kura- wongs.

Additional room volume was gained by pushing the balcony 7m beyond the outline of the historical stables FIGURE 4. Ground level plan (2001). courtyard. This lengthening placed concert patrons at a more favourable distance from the performance platform Well into the documentation process and during ex- and enabled sound and light locks to be introduced at all cavation within the hall, the remnants of a 1790’s bake audience entries. This recessed gallery also freed the house were discovered. This condition sealed the deci- stalls from the acoustic shadow cast by the original bal- sion to replace the original timber floor with a continuous cony. concrete slab mounted on vibration-isolating pads. Choir seating and antiphonal performer opportunities The reinstatement of a resonant courtyard at the cen- were obtained by wrapping a shallow gallery around the tre of the 1820 stables building is the result of the rede- platform on three sides. The organ was fully restored and velopment of the Sydney Conservatorium of Music. Listener envelopment in concert halls and sound from behind

M. Barron

Department of Architecture and Civil Engineering, University of Bath, BATH BA2 7AY, U.K.

It is now generally accepted that there are two audible spatial effects in concert halls: source broadening and listener envelopment (LEV). Bradley and Soulodre have proposed the late lateral level as an objective measure for listener envelopment. But this measure takes no account of sound arriving at the listener from behind. This paper reports on measurements of the front/back ratio in two large concert halls and whether this quantity appears to relate to perception of sound from behind.

The work described here was based on the belief that commented that Bradley’s measure, GLL, is hearing sound from behind is a desirable component of completely independent of sound from behind; though concert hall listening. Two tentative hypotheses were Evjen, Bradley and Norcross [8] have tested that this is involved: 1) that sound from behind is inaudible to subjectively correct. listeners close to or under balcony overhangs and However Hanyu and Kimura [9] and Furuya et al. 2) that hearing sound from behind is related to the [10] find evidence that listeners are influenced by front/back ratio for sound at the listener. The first sound both from directly in front and from behind in hypothesis was based on limited personal observation, their judgement of listener envelopment. Hanyu and while the second could be tested from measurements Kimura consider that a uniform directional distribution in concert halls. The front/back ratio was measured in of sound provides the greatest sense of listener two large British concert auditoria. envelopment, very much in line with the suggestion by Damaske in 1967 [1]. They proposed as a measure for BACKGROUND listener envelopment the spatially balanced centre time, SBTs, though this particular measure omits the Prior to 1967, the spatial aspect of perceived sound level aspect which also influences envelopment was associated with the reverberant sound and the state judgements. of diffusion. Damaske [1] produced a key result: to There are clearly differences of opinion concerning appear surrounded by sound the listener must receive the significance of sound from behind. The work sound from four principal directions in the horizontal discussed here is based on the assumption that the plane, such as front-left, front-right, rear-left and rear- simplest objective measure, the front/back ratio, might right. Discussion of spatial hearing shifted in 1967 to relate to subjective observations. the effects of early lateral reflections, as proposed by Marshall [2]. OBJECTIVE RESULTS Morimoto and Maekawa in 1989 [3] were the first to suggest that there were two components to spatial Measurements were made of the front/back ratio in hearing: source broadening and listener envelopment two large concert halls: the Bridgewater Hall, 3 (LEV). Bradley and Soulodre [4] have conducted Manchester (2400 seats, volume 25 050m ) and the extensive experiments to derive an objective measure Belfast Waterfront Hall (2230 seats, volume 3 relating to LEV: the late lateral level (GLL). The late 30 800m ). The platforms of each hall were empty at lateral level is the level of sound arriving from the side the time of the measurements, which may influence more than 80ms after the direct sound. To detect the the results. Impulses were radiated from an omni- lateral sound, the same technique is used for GLL as directional loudspeaker on stage and consecutive for the early lateral energy fraction, the measure for recordings made of the impulse response with a source broadening. Barron [5] has shown that the microphone first pointing forwards and then major determinant of GLL in actual halls is the total backwards. A hypercardioid directivity was used for acoustic absorption: the measure predicts high degrees the microphone. The front/back ratio was calculated of envelopment in small halls and low degrees in large for three time periods: the early sound (0-80ms), the halls. early reflected sound (5-80ms) and the late sound In 1993 Morimoto and Iida [6,7] suggested that the (>80ms). The frequencies used were the 125 - 1000Hz front/back sound ratio might be significant for concert octaves, except for the early reflected measurement, hall listening. This was the first time serious for which the 250 - 1000Hz octaves were used. Ratios consideration of sound from behind had been are expressed in dB averaged over the frequency discussed. Subsequently several authors have range. Table. Means and standard deviations of front/back ratios in the two concert halls Bridgewater Hall, Manchester Waterfront Hall, Belfast Mean St. Dev. Mean St. Dev. Early front/back ratio (dB) 5.7 2.7 6.0 2.3 Late front/back ratio (dB) 2.3 0.5 2.0 0.8

The two halls are different in design, the 11 Manchester hall is parallel-sided whereas the Belfast 10 hall follows the vineyard terrace scheme. However the Regression line front/back ratios are remarkably similar in the two 8 halls, see Table. The early and late results for the Belfast hall are shown in the Figure. The early front/back ratio decreases from the front to the rear of 6 the hall. We would expect this behaviour due to the direct sound. However the early reflected sound also 4 behaves in a similar way, decreasing as one moves (dB) ratio front/back Early Exposed seats: away from the stage. A possible explanation is that when one is close to the stage, early reflections come 2 Overhung seats: from the front of the hall; when one is towards the rear 1 010203040 of the hall many reflections arrive from behind. Source-receiver distance (m) The behaviour of the late front/back ratio came as a 4 surprise: it is basically constant. Though the constant value indicates uniformity, the mean value of 2dB shows that more reverberant energy arrives from in 2 front. This is not true diffuse behaviour and is, as far as is known, a new result. 0 The original hypothesis was that late sound from the rear might be weak near balcony overhangs. A (dB) ratio front/back Late -2 weak rear sound would produce a high front/back 010203040 ratio. In the figure there is no evidence of this, the Source-receiver distance (m) opposite appears to be more the case with lower values FIGURE. Individual measured front/back ratios in the of the ratio at overhung seats than elsewhere. Waterfront Hall, Belfast. CONCLUSIONS REFERENCES On the evidence of this preliminary exercise, there is 1. P. Damaske, Acustica 19, 199-213 (1967) no indication that the front/back ratio matches the subjective observation of reduced envelopment 2. A.H. Marshall, J. Sound Vib. 5, 100-112 (1967) perceived near balcony overhangs! The perception of 3. M. Morimoto and Z. Maekawa, Proc. 13th ICA, listener envelopment and sound from behind is clearly Belgrade, 2, 215-8 (1989) subtle. The differences of view on listener envelop- 4. J.S. Bradley and G.A. Soulodre, J. Acoust. Soc. Am 97, ment among researchers needs of course to be 2263-71 and 98, 2590-7 (1995) resolved. There is also a major need for subjective 5. M. Barron Applied Acoustics 62, 185-202 (2001) evidence from real concerts of perceived LEV and 6. M. Morimoto and K. Iida, J. Acoust. Soc. Am. 93, 2282 sound from behind. (1993) The measurements reported here provide 7. M. Morimoto, K, Iida and K. Sakagami, Applied interesting new evidence concerning the directional Acoustics 62, 109-124 (2001) distribution of reverberant sound in concert halls, though results from more halls would be welcome. 8. P. Evjen, J.S. Bradley and S.G. Norcross, Applied Acoustics 62, 137-153 (2001) ACKNOWLEDGMENTS 9. T. Hanyu and S. Kimura, Applied Acoustics 62, 155-184 (2001) I am grateful for the help of Dr. J.Y. Jeon during the 10. H. Furuya, K. Fujimoto, C. Young Ji and N. Higa, measurements. Applied Acoustics 62, 125-136 (2001) Investigation of the Factors Most Important for Determining the Acoustic Quality of Concert Halls Y. J. Choi* and F. R. Fricke Department of Architectural and Design Science, University of Sydney, NSW 2006, Australia, *E-mail:[email protected] The purpose of this study is to investigate the acoustic factors that contribute to the overall acoustic quality of concert halls (AQI). The analysis was undertaken using Beranek's six orthogonal parameters (EDT, TI, IACC, Gmid, SDI, and BR) and other factors such as the number of seats and the hall volume. A neural network analysis was used with inputs of Beranek’s parameters over the frequency range 125-1000Hz. Various combinations of acoustic factors were tried to determine which of Beranek’s six parameters are most significant in accurately predicting AQI and what other facotors are important. It is shown that Beranek’s six factors can give good prediction of AQI. It is shown that some other combinations of parameters can give predictions as good as those using Beranek’s parameters.

INTRODUCTION Input Hidden Output Recently, Beranek [1] suggested six acoustical Layer Layer Layer features that must be provided for achieving good EDT acoustics: EDT (Early Decay Time), IACC (Inter-Aural IACC

Cross Correlation), Gmid (the average intensity of the Gmid sound at mid-frequencies), Time to the first reflection TI Acoustic Quality (TI), Bass Ratio (BR) and Surface Diffusivity Index BR Index (AQI) (SDI). Moreover, he indicates how each feature SDI contributes to the overall acoustic quality of a hall and Volume provides the preferred values of six features as follows: Seats IACCE3 of 0.3, Gmid of 4 to 5.5 dB, EDT of 2.2 s, TI of 20ms or less, BR of 1.8 s and SDI of 1.0. FIGURE 1. Diagram of the neural network architecture. Beranek's work is based on Ando's investigation[2] but with two additional factors, BR and SDI. Ando[3] RESULTS expressed concern about the two added factors and Neural Network Analyses were undertaken to their orthogonality. investigate the factors which contribute to the This study is aimed at determining the combination prediction of the acoustic quality of concert halls. of factors required for a good concert hall. As a first Using the data on 20 halls from Beranek's book[1], step, an independent evaluation of Beranek's approach neural networks having different combinations of was undertaken using Neural Network Analysis. Also, inputs were trained over the frequency range 125- a modified version of Beranek's theory that used a 1000Hz. combination of some of Beranek's parameters with Networks having different combinations of acoustic geometrical parameters, was examined to see whether and geometric input parameters were trained. These 17 this might give better results. networks are No.1 (Beranek6+Geo2), No.2 NEURAL NETWORK ANALYSIS (Beranek6+N), No.3 (Beranek6+V), No.4 (Beranek6), No.5 (Beranek6-BR+N), No.6 (Beranek6-BR+V), In the past two decades, Neural Network Analysis No.7 (Beranek6-SDI+N), No.8 (Beranek6-SDI+V), has been extensively studied and applied in solving a No.9 (Ando4+BR), No.10 (Beranek6-TI), No.11 wide variety of problems. NNA is useful for solving (Ando4-TI+SDI,V), No.12 (Ando4-TI+SDI,N), No.13 non-linear problems that are not well suited to (Ando4), No.14(Ando4-TI+SDI), No.15 (Ando4- traditional methods of analysis. In particular, NNA[4] TI+BR), No.16 (Ando4-TI+N) and No.17 (Ando4- is good at pattern recognition and as robust classifiers, TI+V). Before a Neural Network Analysis was with the ability to generalize in making decisions about undertaken, the correlations between the input values imprecise input data. were checked using Statistica software to determine the The following figure shows the neural network orthogonality of the inputs. The inputs are architecture of this research. In the present study, eight approximately orthogonal. The correlations in each inputs and one output were used. A neural network octave are slightly different, EDT and Gmid are the with one hidden layer containing two neurons was most highly correlated for every frequency band. trained. The results are summarized in Table 1, with standard deviation ratios (SDRs), over the seven frequency bands. The SDRs show the degree to which the data has been fitted (A standard deviation ratio of 0.1 or lower indicates very good regression performance). The network models that are shown without their SDRs in Table 1 indicate poor prediction performances. The SDRs indicate quite different trends over each octave band, even though the network has the same input variables. Fig.2-1 AQI as a function of Gmid and EDT Table 1. Standard Deviation Ratios (SDR) for 17 Networks over the four octave bands 125Hz-1000Hz and three combined frequency bands. (No.4 is Beranek's model and No.13 is Ando's model)

Frequency Bands No. 125- 500- 125- 125Hz 250Hz 500Hz 1kHz 250Hz 1kHz 1kHz 1 0.05 0.45 0.21 0.10 0.16 0.09 0.24 2 0.19 0.16 0.11 0.26 0.07 0.23 0.16 3 0.35 0.21 0.52 0.02 - - 0.10 4 0.28 0.17 0.02 0.02 0.14 0.14 0.07 5 0.14 0.09 0.36 0.29 0.31 0.11 0.34 Fig.2-2 AQI as a function of N and EDT 6 0.47 0.14 0.03 0.10 0.38 - 0.13 7------FIGURE 2. AQI Response Surfaces for the best network in 8 - 0.22 - - - 0.20 - the 250Hz using a modified version of Beranek's model. 9--0.21-- -- 10 0.28 0.03 0.07 0.14 0.05 - 0.12 be 2000-2400 though the relationship with the N is 11 0.35 0.09 - - - - - non-linear. 12 0.39 0.05 - 0.10 - - - 13 - - - 0.52 - - - CONCLUSIONS 14 - - 0.25 - - 0.05 - 15------In conclusions, there are several network models, 16------besides Beranek's model which present a good 17------performance to predict AQI. Further more, those According to the results, Beranek's model shows a models consist of some of Beranek's factors together mostly good performance over the each frequency with the number of seats and the volume. This band. In addition, there are several other network indicates a possibility that geometrical factors could be models that gave good prediction of AQI. The model one of the significant parameters as well as the objective parameters to lead to good acoustics quality. having EDT, 1-IACCE3 , Gmid , BR and SDI is good enough to predict AQI and some factors of Beranek's Even though Beranek's model presents one possibility six inputs combined with geometrical data (EDT, 1- to give a good prediction of AQI, there is still a considerable need for further practical investigation of IACCE3 , Gmid , BR, SDI and the number of seats or volume) show good performances as well. Beranek's approach on the preferred values and However, the SDRs do not indicate the interaction weightings. between the inputs. To better understand the ACKNOWLEDGEMENT relationships between inputs, Fig.2 presents the AQI response surface for the factor of Gmid and N on the The authors would like to thank Prof.Y.Ando, best trained network in the 250Hz octave, using a Dr.J.S.Bradley, Prof.A.C.Gade, Dr. T.Hidaka, Prof.G. modified version of Beranek's approach (EDT, 1- Vermeir and Prof. M.Vorlaender for allowing access to IACCE3, Gmid, BR, SDI and the number of seats). the measurement data on specific halls. The results indicate that the low IACC gives the highest AQI, as per Beranek and an EDT of 1.71s REFERENCES would tend to be good. As shown in Fig.2-1, the (1) Beranek, L.L.,Concert Halls and Opera Houses: How they sound, American Institute of Physics, Woodbury, 1996. highest Gmid of 8-9 dB is preferable, while the (2) Ando, Y., Concert hall Acoustics, Springer-Verlag, Berlin, 1985. preferred value of TI of 12 ms is the lowest. The (3) Ando, Y., Architectural Acoustics: Blending Sound Sources, preferred value of SDI would be 1.0. Finally, Fig.2-2 Sound Fields, and Listeners, Springer-Verlag, New York, 1998. indicates that the preferable value of N would seem to (4) Bishop, C., Neural Networks for pattern Recognition, OUP, Oxford, 1995. Analysis and structural adjustment performed to improve the acoustics of the "Strehler Theatre" in Milano G. Zambon and E. Sindoni

Dip.to di Scienze dell'Ambiente e del Territorio, Università degli Studi di Milano-Bicocca, 20126 Milano, Italy

The analysis and the structural adjustment performed to improve the acoustics of the "Strehler Theatre" in Milano is discussed. The theatre was originally designed only for drama and, later on, when the Management decided to perform chamber music and light opera as well, the inadequacy of the acoustic hall soon arose. The Acoustics Laboratory of the Milano-Bicocca University was addressed to solve the very serious problem.

INTRODUCTION especially considering that the lower values are taken when the source is placed in the orchestra pit. In 1997 the new theatre "Strehler" was opened in Milano. The theatre was originally designed only for Sound pressure level (LP ) drama and, later on, when the Management decided to The LP values decrease very fast with distance from perform chamber music and light opera as well, the the source. From the formula of the sound distribution inadequacy of the hall acoustics soon arose. Not only within a room an approximate value of 0.8 for the the concerts and the opera have been heavily criticized absorption coefficient is obtained and therefore the in the news, but many critics complained even about theatre hall is fundamentally sound absorbing. the quality of the acoustics for drama performances. For this reason the Acoustics Laboratory of the FIRST IMPROVEMENTS Milano-Bicocca University was addressed for finding a possible solution to improve the acoustic performance. In order to improve the acoustics of the theatre a series of operations to be carried out gradually was FIRST MEASUREMENTS proposed. The first operation consisted in the replacement of the absorbing material placed on a large Reverberation time (T30 ) portion of the side walls (heavy cloth) with a reflecting The reverberation time averaged over all the material (wood). A second operation was the positions of the theatre hall versus frequency is plotted replacement of the fitted carpet of the stalls with a in Figure 1. wooden parquet. Thanks to these operations the reverberation time improved as shown in Figure 2. 3.5 0.4 3 0.35 2.5 source on stage 0.3 source on pit orchestra 2

(s) 0.25 first operation 30 T 1.5 0.2 second operation

1 0.15

0.5 T30 variation (s) 0.1 0.05 0 125 160 200 250 315 400 500 630 800 1 k 1.25 k 1.6 k 2 k 2.5 k 3.15 k 4 k 5 k 6.3 k 8 k 10 k 0 125 160 200 250 315 400 500 630 800 1 k 1,25 k 1,6 k 2 k 2,5 k 3,15 k 4 k 5 k 6,3 k 8 k 10 k

Frequency (HZ) -0.05 Frequency (Hz) FIGURE 1. Reverberation time averaged over all the FIGURE 2. Increment of reverberation time after the two measurements positions. operations.

It can be noticed that while the T30 values are right An additional result was the improvement in the for drama and opera, they are inadequate for concerts, distribution of LP . ADVANCED MEASUREMENTS grows up, while the stage contribution decreases. A further confirmation of this hypothesis is supported by The first changes improved the acoustic quality of the comparison between the shape of the impulse the theatre, but the result was not yet optimized for response of the nearest positions and the farthest ones. some kind of performances such as musical events. A more accurate characterization of the hall was obtained Inter-aural cross correlation (IACC) by measuring the acoustical parameters: C80, D50, The values of IACC are rather high away from the IACC, ITDG and RASTI in the displayed positions. walls and even higher near the symmetry axis of the stalls. Moving away from the symmetry axis as far as the sector of the theatre where the seats are turned toward the center of the stage, the IACC values decrease. Further there is an increase due to the fact that the signals coming from the stage are again similar (Pos. 06 and 08). From these considerations we can infer that relatively far from the walls the IACC is determined by the direct component of the sound.

Initial time delay gap (ITDG) As expected, this delay, for positions far from the walls, is very long. A detailed analysis of the impulse response shows that Lp of the first reflection is also very weak. FIGURE 3. Position of the receivers. Rapid speech transmission index (RASTI) Table 1 shows the parameters values calculated from The RASTI is badly affected by the reverberation the impulse response. time and is quite high within the whole theatre. It decreases moving away from the source and from the Table 1. Acoustical parameters in the theatre. walls as Lp of the direct and reflected sound are reduced. Pos. D50[%] C80[dB] IACC ITDG[ms] RASTI 01 40.98 0.19 0.82 35.72 0.90 NUMERICAL MODEL 02 37.66 1.41 0.55 19.91 0.86 03 33.85 0.35 0.59 25.72 0.85 The analysis of the results after the structural 04 35.87 1.01 0.42 39.51 0.69 changes performed to date reveals that the acoustics of 05 31.82 2.46 0.54 38.86 0.63 the theatre is far from being satisfactory. This result 06 43.91 2.14 0.63 25.39 0.69 07 44.84 3.50 0.48 31.89 0.65 may be attributed to the quasi total lack of both near 08 57.73 3.57 0.56 17.06 0.67 and far reflections from the sources. Considering the 09 50.94 5.26 0.50 35.74 0.66 structure of the hall and the imposition to avoid any 10 62.83 4.43 0.40 10.51 0.64 substantial architectural change, an improvement is possible by introducing new reflecting surfaces of low visual impact. The first group of such surfaces should Clarity and Definition index (C80, D50) be placed on the sides of the stage and will consist of The analysis of the data shows that the values either pillars resembling roughly half columns cut vertically: for C80 and D50 are low at the first rows but grow to the plane sides will face the stage while the concave unacceptable values for concert and symphonic music sides will face the audience. Two more roughly shell away from the stage and the center of the hall. This shaped reflecting surfaces should hang from the behavior is due to the excessive theatre width, the ceiling: one just above the “boccascena” and the presence of a leaning-out gallery and a poorly orchestra pit to increase the reflections within the reflecting ceiling, whose main effect consists in the stalls, the second hanging from the central part of the severe reduction of the delayed reflections. Near the ceiling to increase the reflection in the balcony. To test stage, where there are no close reflective surface, few the effects of the above changes, an acoustical response early reflections and many delayed ones generated model (RAMSETE) was applied to a CAD simulation from the stage itself are measured. Moving away from of the hall. The results show a substantial improvement the center and the front of the hall the contribution of of all the principal acoustical parameters. early reflections, coming from side and rear walls, Three-dimensional Reverberation-sound Rendering based on Distribution Statistics of Poles and Residues in Transfer Functions M. Toyamaa, M. Kazamab and Y. Kamiyaa aDepartment of Informatics, Kogakuin University,Tokyo, 192-0015 Japan bInstitue of Spatial Science,Waseda University, Japan

A new method for rendering 3-D reverberation–based on the transfer-function statistics in terms of poles and their free oscillations – has been developed. The rendering procedure consists of four steps: (1) modal frequency generation based on a modal spacing histogram; (2) superposition of decaying free-oscillations of modal frequencies with random magnitude and phase; (3) addition of direct sound and early re'ections; and (4) the generation of a pair of binaural impulse responses according to angular distribution statistics of re'ections. Signi`cant parameters of a reverberant sound `eld, such as room volume, reverberation time frequency dependency, modal statistics, initial echoes, and spaciousness can be implemented into this rendering process. As a result the rendered reverberation sound sounds natural.

INTRODUCTION pressed by decaying free-oscillation with random mag- nitude and phase. Arti`cial reverberators have been developed as part of immersive audio technology for computer network com- munication, and such technology requires the modeling STAGE 3: EARLY ECHOES of acoustic events to include three-dimensional(3D) spa- tial sound effects[1]. Reverberation plays a key role in Both direction sound and early echoes can be added 3D effects as well as in sound image localization. Ac- to the generated impulse response. The levels, time gaps, cordingly, we have developed an algorithm for rendering and the frequency characteristics of the early echoes are binaural room impulse responses from transfer function appropriately controlled. The impulse response including statistics, including modal spacing distribution and angu- the early echoes is shown in Fig. 4, and the magnitude of lar distribution. its frequency response is illustrated in Fig. 5.

RENDERING PROCESS STAGE 4: BINAURAL REVERBERATION The rendering process is outlined in Fig. 1. Each of RENDERING rendering stage is described in below. Binaural properties such as spatial effects pose a sig- ni`cant problem for reverberation rendering. We assume that the reverberation sound comes to the binaural receiv- STAGE 1: MODAL FREQUENCY ing position randomly according to the angular distribu- GENERATION tion of sound. An example of a pair of binaural impulse response records is illustrated in Fig. 6 where the distance First, a series of the modal frequencies is generated between the two binaural points is assumed 30 cm[3]. according to the modal spacing histogram[2] shown in Fig. 2. In this stage, only the non-overlapping modes at high modal density are taken. Figure 3 illustrates an example of the resulting series of modal frequencies. SUMMARY

We have developed binaural 3D-reverberation rendering based on random-sound-`eld statistics. STAGE 2: IMPULSE RESPONSE Reverberation-sound is made by superposition of modal- GENERATION oscillation with random magnitude and phase. Room acoustic parameters including reverberation time, early The impulse response is obtained assuming complex echoes, and angular distribution can be implemented. random residues by summing all modal responses ex- Informal listening tests con`rmed natural reverberation sounds. This research was partly supported by TAO, REFERENCES JAPAN. 1. L. Sacioja et alli, J. Audio Eng. Soc., 47, (9), 675-705 (1999) 2. R. Lyon J. Acoust., Soc. Am. 45, 545-565 (1969) 3. M. Tohyama, et al. J. Acoust. Soc. Am., 85, 780-786 (1989)

(1) Modal frequency generation according to Fig. 2

(2) Impulse response generation by superposition of modal oscillations

(3) Adding direct sound and early echoes

(4) Binaural reverberation renderring

FIGURE 1. Rendering Process FIGURE 4. Generated impulse response

FIGURE 2. Modal-frequency-Spacing Histogram FIGURE 5. Magnitude frequency response of the impulse re- sponse shown in Fig. 4

FIGURE 3. A sample of modal-frequency-sequence Reverber- ation Time FIGURE 6. Binaural reverberation impulse response.

Are current room impression criteria suitable for multiple source situations? M. Blaua and V. Ermischa aInstitut für Akustik und Sprachkommunikation, TU Dresden, D–01062 Dresden, Germany

Room impression or spatial impression is considered a major acoustical attribute of concert halls. In order to quantitatively predict this attribute (or its subattributes) for evaluation and design specification purposes, various objective criteria have been proposed over the past 40 years or so, such as Glow, -C80, LF, 1-IACC, and more complex ones. They all have in common that they are based on single-input impulse responses, many of them were developed and optimized in anechoic test rooms where the direct sound emerged from just one loudspeaker. On the other hand, in concert halls one usually does not have just one source (performer) but rather a group of sources (e.g., an orchestra) occupying a certain space on the stage. Preliminary listening experiments conducted in a 16-channel virtual sound field at TU Dresden suggest that with spatially divergent sources fewer early reflections are needed than with monaural direct sound to produce a comparable spatial impression. The results of these experiments and practical observations are discussed together with implications for further research.

INTRODUCTION rooms where the direct sound emerged from just one loudspeaker. Room impression or spatial impression is considered On the other hand, in concert halls one usually does a major acoustical attribute of concert halls. As there are not have just one source (performer) but rather a group of many interpretations of what spatial impression means, sources (e.g., an orchestra) occupying a certain space on we cannot resist to contribute yet another one, based on the stage. This paper is an attempt to explore the impli- the outline given by SCHMIDT [1]: The widely accepted cations of this apparent contradiction. starting point is that a listener unconsciously makes judg- ments about properties of the room he or she is in, as well as about his or her position and that of the sound sources in that room. These unconscious judgments may A PRACTICAL EXAMPLE then give rise to more complex emotional impressions such as “sense of community”, “sense of being enveloped The effect of whether a spatially spread source con- by sound”, “sense of being in one room with the per- tributes to the sound field at the listener’s place or not can impressively be observed at the upper balcony of Dres- formers”, all of which are desired enrichments of musi- 1 cal performances. Spatial impression in the sense that den’s famous opera house . Leaning forward and back acousticians should be concerned about could then be de- just a few centimeters, such that the orchestra is either vis- fined as the result of the acoustical contributions (in ab- ible or not, causes an enormous change in the perceived sence of other contributions such as optical ones) to these acoustical impression. This change is described mainly complex emotions. Note that this definition implies that as a vanishing of the sense of being in the same room acoustical and non-acoustical contributions do not inter- with the performers. act (although non-acoustical contributions may dominate acoustical ones). As this definition illustrates, spatial impression is a A LISTENING EXPERIMENT complex attribute with several sub-components. Histor- ically, reverberance was considered the most important In order to explore the effect of spatially spread sub-component [2]. Today, the most widely accepted in- sources, a preliminary listening experiment was con- terpretation is that of two important sub-components (ap- ducted in a 16-channel virtual sound field at TU Dresden. parent source width and listener envelopment) and further In this admittedly artificial situation, the only means of perhaps less important ones (such as room size, intimacy, changing the spatial extent of the source is to switch be- reverberance seen independently of envelopment, etc.). tween monaural and stereophonic source material. The In order to quantitatively predict either spatial impres- question to be answered was, whether or not it is possible sion as an integral attribute, or its sub-components, var- to create a comparable spatial impression with the two ious objective parameters have been proposed over the source conditions (mono vs. stereo) and how such sub- past 40 years or so, such as Glow, -C80, LF, 1-IACC [3], and more complex ones. They all have in common that they are based on single-input impulse responses. Many 1 Thanks to Dr. Ederer of Akustik Bureau Dresden for providing this of them were developed and optimized in anechoic test example. stereophonic source material need less early reflections 0ms,−1.9dB 23ms,−1.9dB 0ms,−1.9dB than those based on monaural source material, in order to 27ms,−1.9dB create a similar spatial impression, with an average dif- 0ms,0dB 0ms,0dB 43ms,−2.9dB 31ms,−3.0dB ference of 0.15 (IACCE3) and 0.08 (LFE4), respectively. 0°,0° The two other sets of sound fields that were judged 41ms,−1.9dB −30°,0° 21ms,−4.0dB 53ms,−5.0dB −45°,31° 53ms,−4.0dB slightly more dissimilar but still comparable (the stereo-

+10°,39° Reverb,−4.0dB +30°,0° Reverb,−4.0dB phonic fields were then judged to create slightly more +45°,31° spatial impression, whereas the monaural ones created 47ms,−3.9dB 54ms,−4.0dB the impression of a more distant source) had slightly 73ms,−2.8dB 67ms,−2.9dB −85°,0° +85°,0° weaker reflections and were less “spacious”. The differ- ence in objective parameters was similar to that reported

+135°,31° for the first set above.

−135°,31°

Reverb,−4.9dB Reverb,−4.9dB DISCUSSION AND CONCLUSION The observations reported here appear to confirm that 0ms,+2.1dB 25ms,+1.2dB 31ms,−0.8dB 37ms,+1.3dB the spatial spreading of sound sources does influence as- pects of spatial impression. 0°,0° 41ms,+1.6dB −30°,0° 47ms,+1.3dB In the opera hall example, only the direct sound is

Reverb,−4.0dB −45°,31° Reverb,−4.0dB changed. The effect associated with a spatially spread

+10°,39° +30°,0° direct sound appears to be best described by a gain in the +45°,31° “sense of being in one room with the performers”.

61ms,−0.4dB 53ms,+1.5dB In the listening test, both direct sound and reflections −85°,0° +85°,0° were altered. Whereas in the set that was judged the most comparable (fig. 1), the reflections were so strong that the

+135°,31° properties of the direct sound were masked to quite some extent, the direct sound was more dominant for the other −135°,31° two sets, for which again a sense of a distant source began

Reverb,−4.9dB Reverb,−4.9dB to evolve. This suggests that the effect of a spatially spread source cannot be reduced to direct sound alone. Rather, FIGURE 1. Example of two sound fields that were judged it appears to make a difference whether or not reflections highly comparable in terms of spatial impression by 4 subjects. are based on partially incoherent source material. Further Source material was an anechoic recording of the 4th movement research is needed to clarify this issue and its implications of Mendelssohn’s Symphony No.3 in a minor. Top: Sound field based on stereophonic source material. Light background for concert hall acoustics. designates left channel source, dark background right channel In other words, using currently established objective source. Bottom: Sound field based on monaural source mate- parameters means to do something principally wrong, but rial. The reverberation (transparent background) for both sound for the time being we don’t know whether this has impor- fields was created using a software plug-in (StudioVerb by DSP- tant consequences nor how to do better. It may be wise FX) working on stereophonic source material. though to store full impulse response measurements (not just the parameters), just in case. jectively similar sound fields would compare in terms of REFERENCES the known objective parameters. As a result of a long iteration process, three sets of 1. W. Schmidt. Qualitätsbeurteilung von Räumen für Musik- sound fields were found whose spatial impression were darbietungen. In: W. Kraak, W. Schommartz, eds. Ange- judged comparable to a certain extent by 4 subjects. The wandte Akustik Bd. 4. Verlag Technik, Berlin, 1990, set with the highest similarity in spatial impression is de- pp. 110–124. picted in fig. 1. In terms of objective parameters, the 2. W. Reichardt and W. Schmidt. Die hörbaren Stufen des sound field based on monaural material had a LFE4 of Raumeindrucks bei Musik. Acustica 17, 175–178 (1966). 0.16 versus 0.07/0.09 for the left and right parts of the 3. Acoustics - Measurement of reverberation time of sound field based on stereophonic material, respectively. rooms with reference to other acoustical parameters Similarly, IACCE3 was 0.31 for the monaural field versus (ISO 3382:1997). 0.47/0.41 for the left and right parts of the stereophonic sound field. This suggests that sound fields based on Effect of Stage Risers on the Sound of Lower String Instruments Y. Yasudaa and T. Sakumaa aInstitute of Environmental Studies, The University of Tokyo,7-3-1 Hongo, Bunkyo-ku, Tokyo, Japan 113-0033

In order to clarify the effects of stage risers on the sound of lower string instruments, sound fields where a riser is placed on an infinite rigid floor with a point source and with a point force are analyzed numerically. For the analysis, the boundary element method is used for the sound field and the mode expansion method for the elastic board. The effects of the height and sideboards of the riser, the stiffness of the top board and the point force applied to the top on the surrounding sound field are discussed in this paper.

INTRODUCTION sound pressure difference vector between two sides of the board, wp is the displacement vector of the It is often discussed whether stage risers enhance the board, ρ0 is air density, and Q is volume velocity of sound radiation of lower string instruments such as vio- the point source. On the other hand, the application loncellos [1]. However, the acoustic effects of risers and of the mode expansion method with the function its mechanism have not been clarified. It is considered sin(mπx/a) sin(nπy/b) to the equation of motion for the that these effects are caused by the vibration of the ris- board gives the following expression: ers’ boards and by the acoustic reflection from the risers. π π In this paper, we approach the acoustical effects of risers m 2 n 2 2 2 [D(( ) +( ) ) − ω ρph]Wmn = P˜mn + Fmn, (2) through numerical vibro-acoustic coupling analysis for a a b sound field where a violoncello and a riser are placed. where D = Eh3/12(1 − ν2) is the board’s flexural rigid- ity with E the Young’s modulus, ν the Poisson’s ratio, h ρ ˜ NUMERICAL METHOD board thickness, and p board density. Wmn, Pmn and Fmn are expansion coefficients for displacement of the board, sound pressure difference between two sides of the board, z and external force, respectively. Discretization of Eq.(2) point source corresponding to that of Eq.(1) leads to

point force 0.7 wp = B A T p˜ + B A f, (3)

a = 1.2 H ÊÊ π π riser y where T = sin( mi x )sin( ni y )dxdy, A = ij e j a b ij 4δij m jπxi n jπyi π π = ( ) ( ) infinite rigid floor mi 2 ni 2 2 2 , Bij sin sin , ab[D(( ) +( ) ) −ω ρph] a b b = 0.9 a πb π = ( mi xf ) ( ni yf ) x fi F sin a sin b , and F is a point force located at (xf, yf) on the board. Solving Eqs.(1) and (3) gives FIGURE 1. Geometry of the model with a point source, a point sound pressure difference between twosides of the board force and a riser. on each boundary element, and the substitution of the values of elements into the integral equation for the As illustrated in Fig. 1, a riser made of thin boards is sound field gives sound pressures at points in space. placed on an infinite rigid floor with a point source and a point force . In this model, we can replace the infi- nite rigid floor with the mirror image of objects to the floor. By applying normal derivative form of Kirchhoff- Helmholtz integral equation to the replaced model, we can obtain the following equation [2]: 2 Type 0 S p˜ = ω ρ0wp + jωρ0Qd, (1) Type 1 (H = 0.1) Type 2 (H = 0.1) Type 3 (H = 0.2) point force elastic board 2 ÊÊ ∂ ( , )+ ( , ) (G ri rq G rˆi rq ) infinite rigid floor rigid board where Sij = dSq, di = e j ∂ni∂nq ∂(G(rs,ri)+G(rˆs,ri)) exp( jkrp−rq) , G(rp, rq)= , p˜ is the FIGURE 2. Illustration of 4 types of models. ∂ni (4πrp−rq) RESULTS AND DISCUSSION 6 x = 10.0 Four types of models are defined as illustrated in Fig. 2 0 for case study. For the numerical analysis, we defined the -6 amplitude ratio of the point force to the point source as 100 Frequency [Hz] 500 R( f )=F( f )/(− jωρ0Q) and determined R( f ) = 0.2on Normalized SPL [dB] Normalized SPL rigid board the basis of measurement with a real violoncello. Riser’s elastic board without a point force top board stiffness and θ, phase of R( f ), are considered elastic board with a point force (θ = 0) as parameters for the study. In all of the following figures, elastic board with a point force (θ = 3π/2) natural freq. of the board sound pressure level is shown without direct sound, and natural freq. of the cavity receiver points for frequency response functions are fixed on (x, 0.0, 1.2). Characteristics of the top elastic board FIGURE 3. Effects of a point force and top board’s stiffness of 10 2 3 are as follows: E = 10 [N/m ], ρp = 600[kg/m ], h = a riser in Type 2. 0.03[m], and ν = 0.2. Effect of top board’s stiffness and point force. Fig. 6 x = 3.0 3 shows frequency response functions for 4 cases of Type 0

2 normalized by the sound pressure level in Type 0 . It can ] dB be seen that the effect of top board’s stiffness appears at [ -6 6 the frequencies near the natural frequencies of riser’s top x = 5.0 SPL

board and cavity. The point force enhances the magnitude d 0 e of peaks and dips of natural frequencies, and its frequency z -6 mali

( ) r range changes with the phase of R f . 6 x = 10.0 Effect of riser’s sideboards and height. Fig. 4 shows No 0 frequency response functions for 3 types of the risers hav- ing elastic top boards without point forces. In Type 2 and -6 Type 3, where sideboards are placed, there is a wide range 100 Frequency [Hz] 500 of gains approximately from 120 to 300 Hz at all receiver Type 1 Type 2 Type 3 points, and the gains increase with the riser’s height. Fig. 5 shows sound pressure level distributions at 200 Hz for FIGURE 4. Effects of riser’s sideboards and height in 3 types all types with a rigid top board. It can be seen that the of the risers having elastic top boards without a point force. types considerably differ in distributions, and that the ex- istence of the sideboards and the increase in the height 3 Type 0 produce gains in the horizontal direction. 2 30 25 1 35

CONCLUSION 0 3 Type 1 The conclusions are as follows: 1) the stiffness of the 2 35 25 top board and the point force affect the reflected sound 1 20 field only at the frequencies near the natural frequencies 30

[m] 0 of the board, and its frequency range changes with the z 3 30 Type 2 phase difference between the point force and the sound 2 pressure from the point source, 2) the existence of the side 25 1 boards and the increase in the height affect the reflected 35 sound field in a wide range of frequency. 0 3 Type 3 30 2 2025

REFERENCES 1 35

1. L. L. Beranek et al., J. Acoust. Soc. Am., 36, 1247-1262 0 0 1 2 3 4 5 6 7 8 9 10 (1964). x [m] 2. T. Terai, J. Sound Vib., 69 (1), 71-100 (1980). FIGURE 5. Effects of riser’s sideboards and height at 200Hz in 4 types of the risers having rigid top boards. Binaural hearing models for the rating of ASW in room acoustics. J. Beckera,b a Institut für Elektrische Nachrichtentechnik, Aachen University of Technology, 52056 Aachen, Germany b now Ford of Europe, Acoustic centre cologne (ACC), 50725 Köln, Germany

In psychoacoustics different models of hearing have been developed for the localization of sound. With the help of these hearing models several phenomenons in binaural hearing can be explained and evaluated. Different interaural correlation methods (EC, gliding cross correlation ...) have been investigated as one part of these models. This presentation focuses on the results of different binaural hearing models for the rating of ASW. The results of binaural models are discussed with the help of synthetic sound fields which allow to change the impression of ASW infinitely variable. The influence of two different correlation methods are discussed.

INTRODUCTION with an artificial head no additional head related transfer function had to be modulated. The time invariant middle Spaciousness is without doubt an important term for ear transfer function has been modulated according to [5]. characterizing the acoustic quality of a concert hall. It The spectral resolution of the model is given by a filter- is of great interest to find the physical reasons for the bank which mirrors the processing of the inner ear. For amount of spaciousness which a sound field evokes. For this purpose 36 Roex filters are calculated with a spacing the judgement of spaciousness in modern room acoustics of 1 filter / ERB [6]. the quantities ASW and LEV have been established more and more in the last years. In order to predict these quan- tities several high correlating objective measures have been determined. On the one hand the sole search for physically mea- surable quantities which highly correlate with spacious- ness bears the danger of not noticing all parameters which are responsible for this spatial perception. On the other hand the sufficiency of the test signals is not obvious and highly correlation quantities may correlate less for an- other set of test signals. In recent publications a discus- sion arises about IACC as an objective measure for spa- ciousness especially for small rooms, although the human sense of hearing uses a system comparable to the analy- sis of correlation for localization. Models for this corre- lation processing exist already for a long time [1]. These models have been developed and modified in order to ex- plain phenomenons in binaural hearing like trading exper- iments [2] or to distinguish between complex stimuli like π FIGURE 1. Responds of different auditory systems on a hard NoS and NoS0 signals [3].This progress offers the pos- switched noise signal (duration 2 s) binaurally recorded in the sibility to use these models as an objective measure for test room (Decay time 1 s). a: bandpass filter output; b: auditory spaciousness. Good results become more likely due to nerve output (normalized on maximum); c: maximum detection the fact that interaural time delay fluctuations which can after correlation be calculated with these models are said to be responsible for the perception of spaciousness [4]. For transition of mechanical waves to neural pulses a model of auditory nerves was chosen [7]. The limi- tation of the dynamic ranges of this model makes it not HEARING MODEL always applicable. In this case the temporal processing could be described by a system of 5 cascaded adaptation The model used for this research consists of several loops [8]. Figure 1b shows the output of this model on a units. Due to the fact that all recordings have been made hard switched noise signal of 70 dB which was binaurally FIGURE 2. Incident angle of sound sources depending on a: different ϕ values, b: SPL, c: seating positions. Solid line listening test results with standard devation, dashed line time delay determination according [3], dash-dotted line correlation according [2]

recorded in the test room and filtered with the Roex filter A listening studie with these sound fields was made for

o

¨ ¨ ¨ ¨ ¨ a 6 ERB (Figure 1a). ϕ ¢ 0 15 30 45 75 90 (c=100). Before starting the 11 The nonlinear processing leads to high amplitudes in test persons were instructed in the meaning of ASW. Du- the beginning of the signal. This behaviour is essen- ring the test the subjects had to move an acoustic poin- tial for the modeling of the temporal processing of the ter at the left and right border of the sound source. The sense of hearing and it is responsible for several mas- difference of the pointer positions was noticed as the in- king phenomenons. Due to the limitation of human sense cident angle of the sound sources [9]. Figure 2a shows of hearing noise at threshold level is added. The out- the results of the incident angle for the synthetic modi- put of these auditory nerves processors is coupled with fied sound field. Because of the first order fit the model two different algorithms for determination of the inter- results match the listening test results nearly. Due to the aural latencies. The first algorithm contains an extended fact that the hearing models contain nonlinear elements correlation method [2] and the second one a subtraction the models can reflect the dependence of ASW on SPL method of the monaural signals [3]. The interaural time [10]. Figure 2b shows the results for different SPL while delay at the extreme values is marked (Figure 1c) and the Figure 2c shows the results for binaural recording on dif- distribution density function of these marks is calculated ferent positions in the room. The simple adapted models with the mean value and its standard deviation. While the reflect the listing test results well especially for different mean value indicates the position of the sound source in SPL. In further studies the contribution of the different the horizontal plane the standard deviation could be trans- critical bands have to be investigated. formed in an incident angle as a measure for the extension of the sound source. As a first approximation the average of all critical bands was calculated for the evaluation of REFERENCES the incident angle of sound source. 1. Lloyd A. Jeffress, J. Comp. Physiol. Psych., 35-39 (1948) 2. W. Lindemann, J. Acoust. Soc. Am., 80, 1608-1622 (1986) RESULTS 3. Jeroen Breebaart and Steven van der Paar and Armin Kohlrausch, Proc. 16th Int. Congress on Acoustics, ICA98, Since it is not known whether the standard deviation 2, 851-852 (1998) or any other value of the distribution density function re- 4. D. Griesinger, Acoustica, 83, 721-731 (1997) presents the incident angle of a sound source the hearing 5. H. Hudde and A. Engel, Acoustica. 84, 1091-1108 (1998) models have to adapted to listening test results. For this purpose a monaural recording was made in the seminar 6. B.J.C. Moore and B.R. Glasberg, Hearing Research, 28, 209-225 (1987) room of the institute [9]. This monaural recording was

converted into a pseudo binaural recording by a pair of 7. R. Meddis, J. Acoust. Soc. Am., 83(3) 1056-1963, (1988)

¡ filters with the transfer functions Hl f ¡ and Hr f . With 8. D. Püschel, Prinzipen der zeitlichen Analyse beim Hören, o these filters the IACC could be varied from 1 (ϕ ¢ 0 ) to PhD. Thesis, University Göttingen (1988) o 0.17 (ϕ ¢ 90 ). 9. Jörg Becker and Markus Sapp, Applied Acoustics, 62 217-

228 (2001)

¤

¥ π ¡¤¡¤¡

Hl f ¡£¢ cos c log 2 f 10. W. Keet, Proc. 6th Int. Congress on Acoustics, E53-E56

¤

¥ π ¡¤¡§¦ ϕ ¡ Hr f ¡£¢ cos c log 2 f (1968) Spatial Factor of Sound Fields for Gamelan Bali Concert Halls

I G.N. Merthayasaa, I.B. Ardhana Putraa and M. H. Hanzena

aDepartment of Engineering Physics, Institute Of Technology Bandung, Jl. Ganesha 10 Bandung 40132 [email protected] , Indonesia

In order to find out the most suitable conditions of sound fields in Gamelan Bali Concert Hall, a serial investigations has been conducted concerning variation of the spatial and temporal parameters of Gamelan Bali sound fields. The subjective judgement test according to Ando's Theory [1] was performed to a group of subjects by varying the value of Interaural Crosscorrelation (IACC) of sound fields while the temporal factors was kept at the same conditions. Using a piece of oleg tambulilingan as a music signal, the results show that all subjects mostly preferred the small value of IACC. These results confirm Ando's theory, but at some conditions to achieve the same subjective preference, Gamelan Bali Music need more spatial Concert Hall compared with classical music of motif A.

INTRODUCTION out the optimum preferred condition of IACC value for Gamelan Bali music concert hall. The Gamelan Bali Music is already well known as unique Balinese traditional music, it is a percussion EXPERIMENTAL PROCEDURE orchestra music, has been fascinated many Western composer and felt influenced for over a century in In this experiment, beside the IACC value, others European and North American Music. The concerts of parameters value was kept constant. The early delay this music has been perform at many concert hall all time between direct sound with the reflection sound over the world, by many Balinese Gamelan groups – was kept at 20 ms, while the total listening level and also by more than a hundred foreign gamelan received by the subject is kept at 78 dBA. Since we groups based outside of the country. The present paper perform the experiment in a semi anechoic room, we was presented based on the comment from the assume that the reverberation time was also constant. Balinese musician who had experienced to perform Stimuli that simulating actual Gamelan Bali sound their music at some concert halls. Their traditional fields were provided by loudspeakers set up. The loud paradigm state that Gamelan Bali Music is an outdoor speakers set consist of four speakers (I-IV) i.e. the first music based on their acoustical power it could present. one at the frontal of the subject position (I), and the 0 0 And at last their question is: what are the optimum others (II – IV) were placed at an angle of 36 , 54 and 0 objective parameters should be used to design a 90 , respectively, with the frontal speaker. In order to concert hall special for their music? vary the IACC value of the sound field at the subject Throughout an intensive investigation it has been position the sound pressure level of the signal from concluded that all of the significant objective each of the speaker was set, in such a way, that the parameters that describe the sound field in the concert measured IACC [3] in the subject position is shown in hall are the level of listening, the delay time of early Table 1. reflection, the subsequent reverberation time and the interaural crosscorrelation. The spatial factor of the Table 1. Simulation of the IACC value of the Gamelan Bali sound field. sound field is describe by the last parameter i.e. the Signal at Amplitude ratio Measured interaural crosscorrelation (IACC)[1]. Following the Speakers IACC comprehensive study of subjective preference, and the I + IV 1 : 1 0.3 related model of auditory brain system, a theory of the I + II 1 : 1 0.4 acoustics design of a concert hall was developed. In I + II 1 : 0.75 0.5 general, the design solution is known as an inverse I + II 1 : 0.5 0.6 problem, there are many possible shape of the hall I + III 1 : 0.25 0.7 which will satisfy the subjective preference goals of I + II 1 : 0.25 0.8 the design [2]. The present study was intended to find I - 0.9 Table 2 . Significant difference between stimuli. In order to compare the results with the music of IACC 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Motif A [1], the same psychoacoustical testing, i.e. 0.3 X 95% 99% 99% 99% 99% 99% paired comparison methods was applied. As a music 0.4 X 95% 99% 99% 99% 99% signal we chose a piece of music from Oleg 0.5 X 99% 99% 99% 99% 0.6 X - 95% 95% tambulilingan music, which is one of a popular 0.7 X - - orchestra that used as a music for performing the Oleg 0.8 X - tambulilingan dance. Each stimuli, 5second length, 0.9 X which has a certain IACC value, was compared with other stimuli in a pair with a silent of one second. And there is a silent of 5 second between a pair at which the subject gives their response regarding the stimuli they 0,0 preferred. At one session of preference judgment all pair was presented randomly, and the presentation was -0,5 performed 25 times for each subject. The subjects were eight male students between 19 and 21 years old, who -1,0 had a normal hearing ability. The subject's responses were transformed into the scale value of preference by -1,5 using the law of comparative judgment [4]. -2,0 RESULTS AND DISCUSSION Scale Value of Preference 0 0,2 0,4 0,6 0,8 1 I A C C The scale value of preference for each subject was plotted in Figure 1. Figure 2. Comparison between scale value of preference between Gamelan Bali music (solid line) with music Motif A 2,0 (dashed line). 1,5 1,0 For an example, decreasing the IACC value of Gamelan Bali 0,5 up to 0,7 almost does not rising the preference of the audience, at which music Motif A has a scale value of 0,0 preference about almost 50% its optimal value. For the same -0,5 scale value, Gamelan Bali needs the IACC value between 0.4 -1,0 to 0.5. Therefore, we conclude that Gamelan Bali music need more ‘spatial Concert Hall’ compared with classical music of -1,5

Scale Value of Preference motif A. -2,0 0 0,2 0,4 0,6 0,8 1 ACKNOWLEDGEMENTS I A C C Figure 1 . The scale value of preference of all subjects. The author would like to thanks The RUT–V Projects of GoI for their financial support to this The ANOVA was performed to the results in order project. to find out the significant difference of the scale value of preference between the each stimulus, and the REFERENCES results are shown in Table 2. This analysis shows that there is no significant difference between the stimuli 1. Ando, Y., Concert hall acoustics, Springer – Verlag, with the IACC value of 0.7 up to 0.9 and between the New York, Chapt. 1, 3 and 4 (1985). stimuli with the IACC value of 0.6 with 0.7. And there 2. Ando, Y. and Singh, P.K., Global subjective evaluation is a difference at least with 5% significant level for the of sound fields and individual subjective preference for seat pair of stimuli other than mention above. selection, in Music & Concert Hall Acoustics, Conference In order to compare the subjective preference of Proceedings from MCHA 1995, edited by Y. Ando & D. IACC value between Gamelan Bali music with the Noson, Academic Press, London, 1997, pp.39-49. music of Motif A, both data was plotted at the same 3. Nakajima, T., Yoshida, J. and Ando, Y., J. Acoust. Soc. graph as shown in Figure 2. Am. 93, 885-891 (1993). From Figure 2, it is shown that to reach the same 4. Torgerson, W.S., Theory and methods of scaling, John Wiley & sons Inc, New York, Chapt. 3, (1963). scale value of preference, Gamelan Bali music need a smaller IACC value compared with music Motif A.

Effects of Temporal Structure of Reflections and Listening Level on Listener Envelopment

T. Hanyu a, S. Kimurab and K. Hoshib

aDepartment of Construction, Nihon University Junior College, 7-24-1 Narashinodai, Funabashi, Chiba, Japan bDepartment of Architecture, College of Sci. & Tech., Nihon University, 1-8 Surugadai, Kanda, Tokyo, Japan

The authors have proposed the spatially balanced center time (SBTs) [1] as a new objective measure for evaluation of listener envelopment (LEV). The SBTs quantifies the spatial distribution of reflections using the center time Ts for each direction. First we conducted several psychological tests in order to determine the influence of the temporal structure of reflections on LEV. In these tests, LEV changed even though either C80 or reverberation time (RT) was kept constant. Accordingly, the effect of temporal structure of reflections on LEV could not be evaluated by C80 or RT. The Ts and EDT were the measure that correlated well with the psychological scores of LEV in our tests. Next, we conducted tests on the influence of listening level on LEV. The tests of listening level were conducted in different situations of temporal or spatial structure of reflections. As a result, it is clear that the effect of listening level on LEV depends on the temporal and spatial structure of reflections.

INTRODUCTION waveform convolved with a portion of the first movement of a Mozart divertimento about ten seconds Bradley and Soulodre found late lateral energy to be long. The tests were conducted according to Scheffe’s significant for LEV [2]. They proposed the late lateral paired comparison method. Each pair of sound fields sound level LG, for a measure of LEV. As for LG, was randomly presented. Ten subjects rated the LEV because absolute sound level after 80ms is being used, of the second sound field of each pair on a seven-point the influence of direct sound and initial reflected sound scale in comparison with the first one. isn't being taken into consideration at all. Is the LEV really unrelated to direct sound and the relative level First, we examined the influence of direct sound on of reflected sound? Generally, the ratio of early LEV (Test 1). In this test we used five loudspeakers located at the front, ±45 and ±135 degrees relative to energy to late energy, such as C80, is well used in order to evaluate temporal structure of reflected sound. On straight ahead of the listener. The front loudspeaker the other hand, there are other kinds of measures, such radiates only direct sound and other loudspeakers as Ts or EDT, for quantifying the decay of radiate reverberant sound. Five sound fields were used reverberation. Which is the better one to evaluate the in this test. The direct to reflected ratio (D/R) varied effects of temporal structure on LEV? In this study, from –6dB to 6dB at equal intervals of 3dB. Only the four subjective tests were conducted in order to level of direct sound varied so that the absolute level examine the effects of temporal structure of reflections of reflected sound was kept constant. Therefore the and listening level on LEV. listening levels of each sound field became as follows: 71, 72, 73, 75, 77dBA. The reverberation time (RT) of TEST METHOD all the sound fields was 2.2s. Second, we conducted a subjective test in order to A total of four tests were conducted using a simulated determine which measure is the better one to evaluate sound field in an anechoic chamber. We installed 16 the effects of temporal structure on LEV (Test 2). In loudspeakers at a distance of 1.5m from the listeners at this test we used 16 loudspeakers. The front equal intervals of 22.5 degrees in the horizontal plane loudspeaker radiates direct sound and reverberant including the two ears of a listener. When carrying out sound. Other loudspeakers radiate reverberant sound. the tests, we selected the required number of We prepared five sound fields in which reverberation loudspeakers from among the sixteen and reproduced time varied from 1.2s to 3.2s at equal intervals of 0.5s. direct sound or reverberant sound. The directional By expanding time intervals between each reflection response required for reproduction was synthesized after 80ms, the reverberation time was expanded. from the virtual sound source [1]. As a stimulus, we Therefore the C80 and the listening level (72dBA) of used a sound composed of the directional response all sound fields were kept constant.

Next, two subjective tests were conducted in order to Results of Tests 3 and 4 are shown in Figure 2. LEV examine the effects of listening level on LEV. In both increases as the listening level increases, even if the tests the listening level was set to three conditions of sound field does not change completely. The change 60dB, 70dB and 80dB in two different situations. width is dependent on reverberation time (Test 3) or Therefore, six sound fields were prepared in each test. arrival direction of the reflected sound (Test 4). The The first test of listening level was conducted in two interaction in reverberation time and listening level different situations of RT=1.2s and 2.2s (Test 3). In and in arrival direction and listening level were Test 3 we used 8 loudspeakers at equal intervals of 45 statistically significant (p<0.05). These effects cannot degrees in the horizontal plane. The front loudspeaker be explained only by quantifying absolute level like radiates direct sound and reverberant sound. Other LG. loudspeakers radiate reverberant sound. Another test was conducted in two different situations of arrival directions of reverberant sound (Test 4). In Test 4 we 3 3 Y(0.05)=0.20 Y(0.05)=0.40 used three loudspeakers located at the front and ±22.5 2 2 or ±90 degrees. The front loudspeaker radiates only 1 1 direct sound and other loudspeakers radiate 0 0 reverberant sound. In Test 4 the temporal structure of reflections was exactly the same. Therefore the -1 -1 -2 -2 reverberation time of all the sound fields was 2.2s. Test 1 Test 2 -3 LEV of Scale Psychological -3 Psychological Scale of LEV Scale Psychological 0123456 1.2 1.7 2.2 2.7 3.2 RESULTS AND DISCUSSION -9 -6 -3 0 3 6 9 RT(s) D/R (dB) Results of Tests 1 and 2 are shown in Figure 1. The FIGURE 1. Psychological scale of Tests 1 and 2. correlation coefficient between objective measures and LEV psychological scores are shown in Table 1. Table 1. Correlation coefficient between objective Although the absolute level of reflected sound does not measures and LEV psychological scores. **p<0.01 change completely, when the result of Test 1 is Test 1 Test 2 observed, LEV becomes small as the level of direct RT N/A 0.959** sound increases. The values of LG and RT are EDT 0.992** 0.953** constant, because they are unrelated to direct sound Ts 0.977** 0.953** level. Thus, they are not correspondent to the change C80 -0.992** N/A of LEV. On the other hand, the correlations are high LG N/A N/A in the other measures where the value is changed by SBTs 0.972** 0.952** the ratio D/R. This result shows the necessity of considering the effect of direct sound for evaluation of 3 3 Y(0.05)=0.25 Y(0.05)=0.31 LEV. Although the C80 and LG do not change 2 2 completely, when the result of Test 2 is observed, LEV 1 1 becomes large as the reverberation time increases. 0 0 The value of LG is also fixed, because the absolute -1 -1 level of reflected sound after 80ms does not change 2.0s 22.5 -2 1.2s -2 completely either, and it is not correspondent to the Test 3 Test 4 90.0

Psychological Scale of LEV Scale of Psychological -3 change of LEV. On the other measures that quantified LEV of Scale Psychological -3 the decay of the reverberation, a high correlation was 60 70 80 60 70 80 obtained. Ts, EDT and SBTs could obtain the high Listening Level [dB] Listening Level [dB] correlation in both experiments. C80, LG and RT FIGURE 2. Psychological scale of Tests 3 and 4. could not evaluate both experimental results at the same time. These results show that not only the absolute level of reflected sound but also the level relative to direct sound should be considered in order REFERENCES to evaluate LEV. Additionally, it is shown that the temporal structure of reflected sound must be 1. T. Hanyu and S. Kimura., Applied Acoustics 62(2), 155- quantified by both direct sound and the decay of 184(2001). reverberation. Ts and EDT can evaluate both effects reasonably. 2. J. S. Bradley and G. A. Soulodre., J.Acoust.Soc.Am., 97(4), 2263-2271(1995).

Concert Hall Design with Auralization, a Case Study

A. Nagya, A. Kotschyb

aDepartment of Telecommunications, Budapest University of Technology and Economics, Sztoczek u. 2. St.206. Budapest, H-1111 Hungary bR&M Construction and Isolating Ltd., Acoustic Engineering Office, Akna u. 2-4. Budapest, H-1106 Hungary

Room acoustic design is an iterative process. Computer aided acoustical modelling is an efficient instrument to help the design. The aim of this paper is to show the design process of a Hungarian concert hall. We have performed analyses with a ray-tracing based software at every stage of the iteration, determined impulse responses at several receiver points, the distribution of reverbaration time and other acoustical properties. We have also auralized sound samples in order to achieve the optimal sound quality. The concert hall is now under construction. Measurements are to carry out, thus a comparison of the simulated and measured parameters can be done.

INTRODUCTION rays that propagate and are reflected from surfaces. The path of the rays are traced and the detection of Designing a concert hall is an iterative process with them at the receiver points are logged. The advantage lots of compromises. In the last decades computer of RTr is its robustness and that parameter maps are aided design has become an everyday tool also for easily determined. acousticians, which makes this iterative process easier The problem in modelling is that the conditions and faster. During the design of a new concert hall in mentioned above are often hurt. In larger halls the Hungary, we have used a ray-tracing based modelling temperature may change with height; acoustic software to verify each stage of the design process. absorbers and membranes have flexible surfaces and Figure 1 shows the model of the concert hall. We have there are complex structures with sizes smaller than performed several analyses to determine acoustical wavelength, eg. the audience. There are also other parameters of the auditoria. We have also used phenomena, such as diffusion (scattering) and auralization to make the results audible, which let us diffraction that are to take in account. Software consider subjective opinions too. packages differ mainly in the way they handle these problems. Lots of advanced and hybrid methods exist, GEOMETRICAL ACOUSTICS the discussion of them is beyond the scope of this paper. Each room acoustical simulation program has its own modelling method, but the basis of most of these MODELLING ROOMS WITH methods are common: geometrical acoustics. ORCHESTRA, CHOIR AND AUDIENCE Geometrical acoustics is based on three main conditions: the medium of the wave propagation is When designing a concert hall we must not forget homogeneous and quiescent; the boundary surfaces are about modelling the audience and the orchestra (or ideally rigid and the sizes of the room are large choir), for they mean surfaces with high absorption compared to the wavelength. If these conditions are and diffusion. There are several approaches. In our fulfilled we can assume that wave propagation is models we have represented both the audience and the rectilinear and that when a wavefront is reflected from orchestra with simple brick shaped objects. The height a surface the angle of its incidence equals the angle of of these bricks were 1 meter, which is the average the reflection. This allows us to represent wavefronts shoulder height of a sitting person (when modelling as soundrays, and thus to use techniques based on the choir a higher and thinner brick was used). The image source method (ISM) or on ray-tracing (RTr). absorption coefficients were set according to results of ISM works with generating a virtual source for each earlier measurements in which statistical values of reflection path, so the number of these virtual sources absorption in each octave band were determined. The increases rapidly with the reflection order. Beacuse of diffusion coefficients were set as suggested by this and other disadvantages the pure ISM is used only Dalenbäck [1], starting from 20 percent at 63 Hz and for the early reflections. In RTr the sources emit sound increasing with 10 percent per octave. Another question of modelling orchestra is positioning absorption which was essential. The effect of the the source(s). In real life a great part of the emitted changes on RT60 at a receiver point can be seen in sound is scattered by the bodies of the players. To Table 1. With some iteration steps we have reached the model this we have used omnidirectional sources results given in Table 2. placed slightly above the upper plane of the ‘orchestra brick’. Thus the rays going downwards reflect from Table 1. Reverberation times at a receiver point. this plane with high diffusity. Frequency At an early stage At the final stage of design of design 125 Hz 2.88 s 2.25 s 250 Hz 2.65 s 2.22 s 500 Hz 2.09 s 2.02 s 1000 Hz 2.00 s 1.96 s 2000 Hz 1.75 s 1.58 s 4000 Hz 1.42 s 1.40 s

A0 Table 2.The expected acoustic parameters (occupied 01 hall) Parameter Value Mid-reverberation time (RT60)1.99 s EDTaud / EDTstage 0.97 FIGURE 1. Geometrical model of the concert hall Lateral efficiency 0.29 (volume 3900 m3, audience 400 people) Clarity (C80) +2.7 dB Although the clarity values were higher than optimal AURALIZATION the subjective opinions about the auralized samples were good. The differences between each step were Auralization is the process of rendering audible by hardly audible for untrained ears. When comparing modelling the sound field of a source in a space, and the first and the final stage of the design the difference simulating the listening experience at a given position was obvious. in the modeled space ([2]). In order to auralize a sound sample we need binaural impulse responses (BIR) of the examined source-receiver arrangement, with which CONCLUSIONS the sample can be convolved (filtered). There are several types of auralization depending on the sort of In this paper we presented a concert hall design. We have outlined the problems of the modelling process listening (headphone or speaker, 2 or more channels, and have shown an approach for modelling audience etc.). The largest problem in auralization is the speed and orchestra. With the help of auralization subjective of convolution. An interesting solution is given in [3]. judgement was available at each step of the design. The predicted acustical quality of the auditorium is THE DESIGN PROCESS high, which are to be compared to measurements as the concert hall is built. After consultations the first model of the concert hall was created. Several analyses were performed with a ray-tracing based program. We have determined REFERENCES impulse responses for different source-receiver arrangements and auralized sound samples. We have 1. B-I. Dalenbäck, CATT-Acoustic v7.0 User’s Manual, Göteborg, also determined parameter maps of reverberation time 1998 (RT60, RT30), and that of other parameters above the audience. Beside these we have also averaged the 2. M. Kleiner, B-I. Dalenbäck, P. Svensson, Auralization-An Overview, J. Audio Eng. Soc.,1993, pp. 861-874 EDTs above the audience and compared to the averaged RTs and to the EDT expected on the stage as 3. L. Savioja et.al., Creating Interactive Virtual Acoustic suggested by Jordan [4]. Environments, J. Audio Eng. Soc., 1999, pp. 675-705 The results showed high spatial dependency of acoustical parameters and that the reverberation time 4. V. L. Jordan, Acoustical Design of Concert Halls and Theatres. App.Sci.Publ., London, 1980 was too big in the lower octaves. In order to achieve better quality we have placed diffusors along the side walls. These diffusors also gave low frequency The Prediction of Reverberation Time Using Suitable Neural Networks

J. Nannariello and F. Fricke

Department of Architectural and Design Science, University of Sydney, NSW 2006, Australia

A neural network approach to predicting the reverberation time, RT60, at the conceptual design stage of auditoria is presented. As the number of input variables that can be readily identified and quantified at the early design stage is small, the objective of this work is to reduce neural network size, and to obtain suitable or optimal neural networks. The results showed that the generalization performance of neural networks with simplified internal representation is efficient.

INTRODUCTION previously used by Nannariello and Fricke [1]. In the present study, a number of feature selection and Nannariello [1] previously undertook an optimization algorithms supported by the neural investigation of the efficacy of neural networks to network software model, STATISTICA Neural predict the low frequency and mid frequency Networks [5] were used to detect interdependencies reverberation times, RT60(125-250) and RT60(500-1000) between input variables. The designated input variables respectively for unoccupied auditoria. The results of the used in the present study represents the same simplistic previous work indicated that neural networks trained information available to an architect and acoustical with as many as 15 input variables provided accurate consultant at the conceptual design stage. Optimal low-band RT60(125-250) and mid-band RT60(500-1000) neural networks were used to make low-band and mid- predictions for the tested enclosures. One of the band reverberation time predictions and to show objectives of the present work is to investigate the relationships between the input variables and the efficacy of neural networks (NN) presented with an predicted reverberation time. optimal data set acquired by way of a dimensionality reduction, i.e. a reduction in the number of input RESULTS USING OPTIMAL N-NETS variables, thus reducing the network size and simplifying internal representation [2, 3]. The present A series of neural networks with various set-up work draws on the results from Nannariello [1] and functions, consisting of several combinations of the attempts are made to reduce network size and to obtain designated set of input variables, were trained, and optimal neural networks. The results of investigations verified. Table 1 shows the results of thirty-four neural carried out to ascertain the efficacy of these optimal networks trained with seventeen different set-up networks to predict low-band RT60(125-250) and mid-band functions for each of the low-band and mid-band RT60(500-1000) for concert halls, auditoria, and churches frequency reverberation times. The table shows that an are presented. Another objective of the present work is eight input variable neural network model (SF7) to interpret the meaning of network size reductions by produced the best results for both low-band and mid- way of investigating: 1) the significance of pruning and band frequencies; with an R2,(RMS error),[SDR] of combining geometrical and descriptive input variables, 0.91,(0.20),[0.34] and 0.97,(0.13), [0.31] respectively. and 2) the non-linearity of relationships between the Tables 2 and 3 show the results of low-band and mid- input variables and the predicted reverberation time. band reverberation time predictions made by the optimal neural networks for six enclosures respectively. INPUT DATA AND PROCEDURE Fig. 1 shows RT60(125-250) as a function of tube ratio, TR, and the subjective enclosure rating coefficient, EAR, The neural network analyses were carried out using and illustrates the non-linearity which exists between a transformed [4] and contracted version of the data some of the input variables and reverberation time. Table 1. Correlation coefficients (R2), Root Mean Squared errors (RMS), and Standard Deviation Ratios (SDR) for neural networks trained with different set-up for low-band (RT60(125-250 Hz)) and mid-band (RT60(500-1000 Hz)) predictions.

RT60(125-250 Hz) RT60(500-1000 Hz) Set-up function Input variables R2 RMS SDR R2 RMS SDR SF1 f (RV, CAS) 0.08 0.58 0.96 0.83 0.31 0.47 SF2 f(RV, CAS, EAR) 0.78 0.28 0.46 0.87 0.44 0.71 SF3 f(RV, TR, CAS, EAR) 0.37 0.49 0.81 0.64 0.46 0.74 SF4 f(RV, TR, CAS, TSA, EAR) 0.63 0.48 0.72 0.85 0.44 0.67 SF5 f(RV, TR, CAS, TSA, AFA, EAR) 0.60 0.48 0.70 0.92 0.34 0.50 SF6 f(RV, MAH, TR, CAS, TSA, AFA, EAR) 0.85 0.26 0.43 0.80 0.34 0.55 SF7 f(RV, MAW, MAH, TR, CAS, TSA, AFA, EAR) 0.91 0.20 0.34 0.97 0.13 0.31 SF8 f(RV, MAH, CAS) 0.08 0.58 0.96 0.70 0.47 0.74 SF9 f(RV, MAH, TR, EAR) 0.47 0.46 0.76 0.85 0.31 0.50 SF10 f(RV, MAH, TR, CAS) 0.11 0.58 0.95 0.80 0.32 0.51 SF11 f(RV, MAH, CAS, EAR) 0.58 0.42 0.67 0.84 0.43 0.68 SF12 f(RV, MAH, TR, CAS, EAR) 0.86 0.24 0.41 0.80 0.32 0.51 SF13 f(RV, MAH, CAS, TSA, AFA, EAR) 0.77 0.31 0.49 0.78 0.43 0.69 SF14 f(RV, MAW, TR, CAS, TSA, AFA, EAR) 0.79 0.29 0.48 0.92 0.27 0.41 SF15 f(RV, MAH) 0.06 0.59 0.97 0.70 0.44 0.70 SF16 f(RV, MAW, MAH, TR, CAS, TSA, AFA) 0.13 0.57 0.93 0.80 0.30 0.48 SF17 f(RV, MAW, MAH, TR, CAS, AFA) 0.16 0.57 0.93 0.80 0.31 0.50

Table 2. Results of SF7 for RT60(125-250) surface plots have shown interesting trends of the relationships between the input variables and the a b c Auditorium MRT60(125-250) PRT60(125-250) %A predicted reverberation time. 1 2.10 2.02 96 2 2.30 2.04 89 3 2.40 2.63 94 4 2.48 2.14 86 5 1.60 1.70 94 6 3.60 3.67 98

Table 3. Results of SF7 for RT60(500-1000)

a b c Auditorium MRT60(125-250) PRT60(125-250) %A 1 0.80 0.79 99 2 1.80 2.00 90 3 2.30 2.48 93 4 2.65 2.82 94 5 2.40 2.33 97 6 2.40 2.37 99 FIGURE 1. Surface plot of RT60(125-250) as a function of a -1 MRT60 = measured reverberation times (sec.) tube ratio, TR, (m ) and subjective rating coefficient, EA b R PRT 60 = predicted reverberation times (sec.) c %A = % agreement (MRT60 and PRT 60) REFERENCES

1. Nannariello, J., and Fricke, F.R., Applied Acoustics, 58, CONCLUSIONS 305-325 (1999). 2. Stein, R., AI Expert, 8(2), 42-47 (1993). Generally, the low-band and mid-band RT60 prediction 3. Mozer, M.C., and Smolensky, P., Connection Science, accuracy of the optimal neural network models, for the 1(1), 3-16 (1989). six enclosures tested, is within the range of the 4. Stein, R., AI Expert, March 1993, 32-37 (1993). subjective difference limen (
RT/RT W 5%) where RT = 5. STATISTICA Neural Networks, StatSoft Inc, Tulsa, OK, reverberation time in seconds. A number of response (1999). The Effect of Location of the Sound Source upon the Subjective Acoustic Indicators of a Hall V. J. Stauskis

Vilnius Gediminas Technical University, Traku 1/26, LT – 2001, Vilnius, Lithuania

The position of the sound sources influences strongly the listeners seated at the front of the hall, whereas the listeners in the back rows are almost unaffected. The larger the distance from the orchestra, the smaller the difference. As the distance between the orchestra and the listener is increased, the values of the C80 index decrease from 0 dB in the front rows to –8 dB in the back rows. When the organ is playing, the change in the index values is not so significant: from –8 to –12 dB. The largest difference in C50 and C80 indexes is seen in the front rows, with large negative values. In the middle and back rows, the values of this different become positive and almost stable. When the orchestra is playing, the variation of the low frequency index C80 (3L) is slight in the front rows is slight but they decrease sharply as the distance from the sound source S1 becomes larger. In case of organ music the values of the index are only negative and are almost not dependent on the listeners’ position.

The investigations were conducted in St. Johns’ (Šv. The sound source S1 is on the floor plane and S2 is -RQÐ FKXUFK LQ 9LOQLXV ,WV OHQJWK WR WKH DOWDU LV  P 7.8 m above the floor plane. Therefore, the distances width at the altar is 24.5 m, and the maximum height is from each measurement point to the sound sources S1 20.4 m. The volume of the hall is 27,000 m3. and S2 will be different. These distances are presented in Table 1. Table 1 Sound source Measurement points and distances to the sound sources, m 1 357891011 S1 3 3.57.51520253035 S2 41 36 31 23 19 14 10 6.5

THE DEPENDENCE OF THE The graph shows that when the orchestra is playing SUBJECTIVE ACOUSTIC INDICATORS (sound source S1) and the distance between the listener and the sound source becomes larger, the value of C80 OF A NON-FILTERED SIGNAL UPON changes strongly: it decreases from 0 dB at the front THE POSITION OF THE SOUND rows to –10 dB at the back rows. SOURCE When the sound source S2 is located at the back of the hall and the organ is playing, the changes in the The evaluation of the hall acoustics by the listener is value of C80 are not significant: from –9 dB at the front always subjective. Subjective indicators are connected rows to –7(8) dB at the middle and back rows. This with the relationship between the early energy and the means that when the orchestra is playing the early late energy. A smaller interval of early energy is taken energy prevails in the front rows, while the late energy for speech, and larger – for music. The energy ratio prevails under the organ music in the same rows. As changes along with the change in the position of the the organ is playing, only late energy formed by the sound source. Fig. 1 depicts the change of C50 and C80 multiple reflections from all planes prevails in all rows. under the action of S1 and S2 at all points of Fig. 2 shows the changes in C80 at various rows, with measurement in the hall. the separate action of sound sources S1 and S2.

                  ,QGH[ & &   

 &ODULW\ LQGH[ &    6  6  6  6  6  6  6  6  6  6  6  6  6  6  6  6 0HDVXUHPHQWV SRLQWV VRUFHV QXPEHU 0HDVXUHPHQWV SRLQWV

FIGURE 1. The Dependence of the Speech Clarity Index C50 FIGURE 2. The Dependence of the Music Clarity Index C80 and the Music Clarity Index C80 upon the Position of the upon the Position of Sound Sources S1 and S2. 1 – S1 is Listener and the Sound Source. 1 - index C50; 2 – index C80. active; 2 – S2 is active. When the orchestra is playing (source S1), the value increase in the distance from S1. In case of organ of C80 strongly depends on the listener’s position. music, this energy is low in all rows. When the distance to the source is small, the index Fig. 4 depicts the change in the low frequency index values are equal to 0.5-0.6 dB. In the middle rows, a C80(3L) which was obtained by averaging out the low sudden drop to –9 dB is observed, whereas no and medium octave frequencies of 125, 250 and 500 significant change is seen in the back rows. Hz. When the organ is playing (sound source S2), there is   little change in the C80 index: from –7.1 to – 9.5 dB.  This means that, in case of orchestra music, the   listeners in the front rows are most affected by the   early energy which is determined by strong direct  sound and several early reflections. In case of organs, 

&ODULW\ LQGH[ & /  G% all listeners irrespective of location are almost equally  affected by the late energy determined by the long 0HDVXUHPHQWV SRLQWV reverberation time. In a hall filled with listeners the FIGURE 4. The Dependence of the Low Frequency Music reverberation time will be shorter but the regularity Sound Clarity Index C80(3L) on the Position of Sound will remain the same. Source. 1 – sound source S1; sound source S2.

The Dependence of Subjective Acoustic When the orchestra is playing (sound source S1), the Indicators on Frequency and Position of values of C80(3L) are quite stable in the front rows (around 0). As the distance from the sound source the Sound Source increases, these values are equal to –10 –11 dB at the back of the hall. When the organ is playing (sound Any work of music is perceived by the listener as a source S2), the values of the index are only negative great deal of sounds whose frequencies and energy and are almost not dependent upon the position of the varies within very broad limits. So far the changes in listener. indexes were examined for an unfiltered signal. Fig. 3 presents the average results of investigations of the Conclusions C80(3) index at the frequencies of 500, 1000 and 2000 Hz. 1. The position of the sound sources influences   strongly the listeners seated at the front of the hall,   whereas the listeners in the back rows are almost  unaffected. The larger the distance from the orchestra,   the smaller the difference.  

&ODULW\ LQGH[ &  2. When the orchestra is playing, the variation of the   low frequency index C80 (3L) is slight in the front rows 0HDVXUHPHQWV SRLQWV is slight but they decrease sharply as the distance from the sound source S1 becomes larger. In case of organ FIGURE 3. The Dependence of the Music Sound Clarity music the values of the index are only negative and are Index C80(3) on Frequencies and Position of Sound Source. 1 – sound source S1; sound source S2. almost not dependent on the listeners’ position.

When the orchestra is playing, i.e. where the sound REFERENCES source is in the position S1, positive values of the index (0.6-0.7 dB) are recorded in the performers’ area. 1. Bradley J. S. Experience with New Auditorium Acoustic Measurements. // J. Acoust. Soc. Amer. As the distance from the sound source increases, C80(3) index decreases suddenly to the values from –9.5 to – Vol. 73, 2051–2058 (1983). 11 dB in the middle and at the back of the hall. When 2. Bradley J. S. Auditorium Acoustics measures from the organ is playing, i.e. with the sound source S2 Pistol Shots. // J. Acoust. Soc. Amer. Vol. 80(1), July , 199–205 (1986). active, the C80(3) index is only slightly dependent on the distance and its values are negative, varying from – 3. Stauskis V. J. Acoustic of Musics halls. Vilnius, 9.5 to –11 dB. This means that, as the orchestra is 1999, 412 p. playing, the high frequency energy prevails in the 4. ISO 3382: 1997 (E). Acoustics – Measurement of performers’ area; it decays significantly with the the reverberation time of rooms with reference to other acoustical parameters. La Fenice - acoustical planning in the scope of a historical reconstruction G. Müllera and J. Reinholda aMüller-BBM GmbH, Robert-Koch-Str. 11, 82152 Planegg, Germany

The famous Teatro La Fenice in Venice was built in 1792. This typical Italian opera house burnt down in 1836 and again in 1996. Opera enthusiasts, musicians, critics as well as town officials demanded a precise reconstruction. They expected that thereby the well accepted "acoustics" of the famous opera could be achieved again. However, when starting any rebuilding, immediately modern requirements on security, stage machinery, air-condition and enlarging of the orchestra pit arise. The relatively high demands on the room acoustical quality of the house, the extremely limited space of the whole building complex in connection with the requirement of an increased number of rooms require a sophisticated room and building acoustical planning and realization.

REQUIREMENTS SET BY THE ROOM ACOUSTICAL PLANNING COMPETITION A complete reconstruction according to the historical Already in 1996, i. e. shortly after the destruction of the model of the audience room, the stage including the orches- theatre, the city of Venice put a very detailed competition tra pit with the aim of achieving the originally good room out to tender. The competition regulations stated that the acoustical quality leads to the question in how far a "room audience area and stage area should be reconstructed acoustical planning" is necessary at all. If it was necessary completely according to the historical model by using the to follow this goal exclusively based on all construction original materials and applying historical working tech- details and construction elements, one could come to the niques in order to restore the excellent acoustical conditions. conclusion that this task can be fulfilled by an architect A total reconstruction is also required for the Sale only. Special safety requirements and particularly those Apollinee, which includes the foyer and entrance zones and concerning fire protection make it, however, necessary to also some rehearsal rooms. replace the original combustible materials at some places by For the wings adjoining in the north and in the south, in fire-resistant ones. In this connection, the acoustical ad- which the side stage, several rehearsal rooms, changing- equateness must be achieved by applying properly chosen rooms, and cloakrooms, workshops, and the administrative materials. Sometimes this must also be proven by acoustical rooms are situated, more freedom was granted although the measurements. partly existing outer shell of the building had to be The acoustically relevant boundary surfaces such as the respected. ceiling of the hall as well as the ceilings of the boxes will be In the scope of the strict requirements for the reconstruction, reconstructed with the original special plaster which is char- modern safety standards - particularly those concerning fire acterized by a constantly varying material thickness and protection as well as the improvement of the air condi- bending stiffness. In connection with the sprinkler system of tioning and the extension of the room programme had to be the ceiling cavity above, a system meeting today's require- taken into account. ments for fire protection can be chosen. According to the competition, the technical rooms, which The same is valid of the gallery balustrades, which can be are necessary for this purpose and further room produced with completely reflecting wooden surfaces, requirements had to be integrated in the building. which comply with modern safety standards. At the same time, the demands for an enlargement of the The relatively short reverberation times which are typical volume of the orchestra pit, an equipment of the stage with for Italian theatres with galleries should be increased as far the latest technique and an increase of the number of audi- as possible. This is achieved by reproducing the fabric ence seats without impairing the room acoustical conditions curtains which have an exclusively decorating effect out of were established. gypsum and sticking fabric on them.

The required enlargement of the orchestra pit can be Due to the limited space the ventilation control had to be realized only below the stage by maintaining the front edge situated above the audience area. Because of fire protection of the orchestra pit and the lateral boundary surfaces. A and acoustical reasons, a reinforced concrete ceiling was movable rear wall of the pit allows a rear boundary as in the necessary in contrast with the historical concept. historical model. In case of large orchestras, a simple, vari- As between the ceiling of the hall and the new ceiling to able adjustment will be possible. the technical control room there is an air cavity of Minor modifications at the balustrade of the orchestra pit approx. 2 m, no influence on the room acoustical quality of as well as the realization of portal reflectors below the the audience room has to be expected. lighting bridge lead to an improvement of the contact For the Sale Apollinee a certain number of different types between musicians in the pitch and singers on stage and of wooden beam ceilings have to be reconstructed. The vice versa. different room separations partly require very expensive For concerts, in the stage room a concert shell will be in- ceiling constructions. To make matters even more stalled. Thus, an optimal acoustical contact between the complicated, all room areas have to be supplied with air musicians and an acoustical coupling as well as an effective conditioning in contrast to the historical concept. The enlargement of the room acoustically effective volume will necessary cable feedthroughs were mainly done in invisible be achieved. It will be possible to increase the slits in the area of the outside walls and in the ceiling cavity. reverberation times for concerts compared with the The close neighbourhood which is typical for Venice reverberation times for opera performances. requires an utmost caution for the planning of air inlet and outlet openings and for the mounting of technical devices inside the building. As a Venetian distinctive feature, the ship traffic on the adjacent channels and on the bridges in the neighbourhood must be taken into account. Big transportation cranes and the relatively short distance from the façades lead to relatively high acoustical requirements for the exterior construction elements.

BUILDING ACOUSTICAL PLANNING The wings adjoining the audience area in the north and in the south, can be separated effectively by an acoustical joint.

A very high noise protection standard can be achieved also for the new rehearsal halls which will be situated below the audience area. A less effective acoustical separation is, however, possible between the corridor zones and the audience hall which is surrounded by them. The reconstruction requires a construction with wooden beams made out of terrazzo which is typical of Venice. Terrazzo results in a relatively low noise protection which is typical of historical opera houses. The single doors to the boxes will confirm the originally existing low noise protection.

Difference Limen for Level of Music

W. J. Davies

University of Salford, School of Acoustics and Electronic Engineering, Salford M5 4WT, UK. [email protected]

A psychoacoustic experiment has been conducted to determine the difference limen (DL) of the level of music reproduced in a listening room. The effects of three factors on the DL were measured using up to twelve trained subjects. The factors were music motif (Mozart, Elgar, Smashing Pumpkins), listening level (50, 60, 65, 70, 80 dB, A-weighted Leq) and dynamic range. It was hypothesised that the DL would reduce with reduced dynamic range or increased listening level. However, both hypotheses are rejected: listening level was the only significant factor, with a mid-range level of 65 dB giving the lowest mean DL of 1.5 dB. This rises to a mean DL of 2.1 dB at listening levels of 70 and 80 dB. These figures are larger than has previously been assumed and are about the same size as the spatial standard deviation in strength across classical concert halls. The results should be of use to designers of concert halls, recording studios and other spaces for critical listening to music.

INTRODUCTION might be—from one place in a room to another, or from one room to another, or from one design to When a sound source operates in a room, the sound another, for example. pressure level will vary across the room. If the room is intended for critical listening (an auditorium or a listening room, for example) then the room designer METHOD will usually seek to minimise or at least control this variation. What should be the target for controlling the The music was reproduced to subjects through stereo spatial variation of level? What would be perceived as loudspeakers in a listening room. An efficient a ‘large’ change in level? One way of answering adaptive psychometric method was used to conduct the these questions is to look for the smallest change in test [3]. The stimulus is played to the subject at two level which can be perceived: the difference limen different listening levels and their task is to say (DL). This has been measured many times using whether they perceive a difference in level. The test simple signals like sine waves, but not, so far, with was controlled by a computer program that varied the music as the source signal. The data for single- difference between the two levels in each presentation frequency measurements show quite a large variance according to the subject's last response. This allows it across different experiments. Luce and Green [1] to approach the subject’s DL to within a pre- plotted results from six different studies showing determined level of accuracy (0.5 dB) in the minimum difference limen (DL) against sensation level (SL) for number of presentations. a 1 kHz tone. Generally, DL decreased as SL Three parameters were varied, one at a time: music increased, but the change was not monotonic in all motif, listening (sensation) level and dynamic range. experiments. The mean DL was 0.9 dB at 50 dB SL, It was hypothesised that the DL may vary with music and 0.6 dB at 70 dB SL. motif (perhaps across genres), that it would decrease Generally, smaller DLs are produced for as listening level increases, and that music with a continuous signals than for transients. Music is a smaller dynamic range would produce a smaller DL. much more complicated signal than a sine wave, so we Three different musical motifs were used: Mozart should expect changes in its level to be harder to (Horn Concerto: Rondeau: Allegro vivace), Elgar judge. In the architectural acoustics literature, various (Enigma Variations: Nimrod) and Smashing Pumpkins values for the DL with music have been assumed, (Mellon Collie: Thru the eyes of Ruby). When the though 1 dB seems to be typical [2]. The work motif was varied, the listening level was fixed at 70 dB reported here seeks to provide for the first time a (A-weighted Leq) at the listening position and the difference limen measured with music in a realistic music was uncompressed. Five different listening room acoustic. The results will be of interest to room levels were used: 50, 60, 65, 70 and 80 dB (A- acoustic designers who would like to know how weighted Leq at the listening position). When the level audible a measured or predicted change in sound level was varied the motif was fixed at Mozart and the music was uncompressed. The effect of dynamic 5 range was measured by electronically compressing the music and characterised by the difference L1-L99, 4 measured at the listening position. The values used were 16 (uncompressed), 13.5, 12, 9.5 and 8 dB (all 3 Mozart at 65 dB listening level). All the subjects were acousticians or acoustics 2 students, all had normal hearing at standard DL (dB) audiometric frequencies and all were experienced participants in psychoacoustic tests. Each subject 1 completed a training programme, consisting of four consecutive measurements of their DL for a fixed 0 combination of the parameters. Twelve subjects began 50 60 70 80 the training programme; between four and seven Listening level (dB) subjects completed all tests for each parameter. FIGURE 1. DL against level for music (upper line, ± one standard deviation); and for 1 kHz sine wave (lower lines: RESULTS AND DISCUSSION minimum, mean and maximum, after [1]).

Three separate two-way analyses of variance the commonly assumed DL for music of 1 dB, and the (ANOVA) were conducted. Each ANOVA examined mean DL for a sine wave. For all listening levels the variance due to one parameter compared with the except 65 dB, the DL is also greater than or equal to variance between subjects. The results are the spatial standard deviation of strength G measured summarised in Table 1. It is clear that the only factor in three halls by Bradley [4]. This suggests that, for significant at the 10% level was listening level. Bradley’s halls at least, judgements of sound quality There was some evidence of interaction between between seats would be made primarily on criteria subjects and each of the factors. An interaction other than sound level. between music motif and subject, for example, implies that some subjects found it easier to hear changes in level in one motif, while other subjects found another CONCLUSIONS motif easier. There is also tentative evidence that more complex music produces a larger DL. The rock The difference limen for the sound pressure level of music track gave 1.4 dB, as against 2.5 dB for Elgar music has been measured. It was found to vary and 2.6 dB for Mozart. However, these effects are not significantly with listening level, but the musical motif significant at the 10% level. and its dynamic range were not significant. The size Perhaps surprisingly, the hypothesis that dynamic of the limen is significantly larger than that previously compression would reduce the DL is also rejected. assumed. Future work might reveal whether it varies This implies that it is not simply the range between the with other factors, such as reverberation. minimum and maximum levels in music that controls its amplitude perception. Finer detail, like the rate of amplitude change, as well as the rhythmic and tonal complexity, is likely to be more important. ACKNOWLEDGMENTS Because level is the only significant factor here, we proceed to average across the other factors to The author is grateful to Mr David Ward for produce the graph of DL against level in Fig. 1. The conducting the tests and to Dr Trevor Cox for writing mean DL is seen to be considerably larger than both the test program.

Table 1. ANOVA results. REFERENCES Factor n P-value Music motif 3 0.21 1. Luce, R.D. and Green, D.W., J. Acoust. Soc. Am., 56, Subject (music tests) 7 0.74 1554-1564 (1974). Dynamic range 5 0.19 2. Barron, M., J. Acoust. Soc. Am., 98, 2580-2589 (1995 ). Subject (range tests) 4 0.31 3. Taylor, M.M. et al., J. Acoust. Soc. Am., 74, 1367-1374 Listening level 5 0.06 (1983 ). Subject (level tests) 5 0.12 4. Bradley, J.S., J. Acoust. Soc. Am., 89, 1176-1192 (1991). Using Room Acoustic Model Tools to Determine Surface Diffusion Effectiveness R.H. Campbell

ECE Department, Worcester Polytechnic Institute, Worcester, MA 01609, USA [email protected]

Room acoustic computer modeling continues to improve with respect to both frequency-sensitive surface diffusion simulation and also the ability to more accurately auralize the space. In the former case, a new Audio Engineering Society Information Document [1] represents the first step in the standardization process to establish a surface diffusion measurement standard with useful coeffi- cients. In the latter case, the auralization component allows placement of directional sources and receivers in the space. Several computational tools are used together in a room acoustic model to determine the spatial effectiveness of modeling surface diffusion. This paper discusses a simple rectangular room, h=8m, w=10m, l=24m, which is modeled with a range from specular to 100% dif- fuse surfaces.

behavior surrounding a particular receiver. The issue INTRODUCTION here is to focus on the minimal combination of output variables that relate strongly to room model surface Room acoustic computer models that incorporate sur- diffusion. face diffusion parameters offer an opportunity to ex- Programs such as CATT can generate a great deal of amine the acoustic behavior in a model room where detailed output data in the process of predicting room these parameters are varied over a wide range. In addi- acoustic parameters. The echogram is one example, tion, receiver data output tools are available to interpret where five types of data are superimposed in the image the results, such as auralization and graphical images of in various colors. If one is interested only in diffuse receiver energy arrival in 3-D space. For the present reflections, it is easy to see them in red, but to print state of computer modeling capability, an understand- them here requires color separating the image. ing of the influence of model surface diffusion values, In Fig. 1, top is the echogram with no diffusion, bot- with respect to prediction results, indicates the level of tom has 100% surface diffusion. There is, clearly, a accuracy required in valuing the surface diffusion pa- high density of diffuse reflections in the bottom trace rameters. This in turn leads to the level of accuracy up to the pre-determined truncation time. required to measure and report the parameters on real physical surface treatments.

THE MODELING PROGRAM

The computer program used in this study is CATT- Acoustic [2], which has the ability to utilize stated sur- face diffusion parameters in the range from 0% to 100%. This program uses a "randomized tail-corrected cone-tracing" technique as a general algorithm for room acoustic prediction. A surface tagged as "diffuse" ac- quires a multiplicity of patches that, when excited, act as elementary sources radiating according to Lambert's law [3].

Available Tools

One can view the acoustic activity in the room model in many ways. Aside from the common measures, such FIGURE 1. Upper echogram with 0% surface diffusion, as LEF, RT, EDT, IACC, STI, C80, etc., there are lower with 100% surface diffusion, 2KHz octave band fil- sound roses that depict 3-D directional energy arrival tered. The two line traces are Schroeder integral (upper), and and also post-processing impulse response and aurali- 2ms energy moving average (lower). The total length of each trace is 250 ms. zation files that can reveal much about the time-domain Sound roses depict, in the X-Y-Z directions, the echogram arrivals displayed over user-determined time periods. One can program the time periods as accumu- lative ( 0 to X) or as discreet (X to Y). In Fig. 2, the top image is for no surface diffusion on all wall surfaces, the bottom is for 100% diffusion, filtered in the 500Hz octave band for approximately the first mean free path. The proliferation of diffuse reflections in the bottom figure can be assessed with respect to direction of arri- val.

FIGURE 3. Typical B-format WXYZ receiver impulse re- sponses in the model room with 15% uniform absorption and 100% uniform diffusion. The omnidirectional trace is W, X is transverse, Y is longitudinal and Z is vertical.

Finally, the source can be setup as a highly directional sound emitter and "shone" about the room as a torch, while auralizing to a fixed omnidirectional receiver. If air absorption is turned off, then each octave band re- acts the same way to the ray tracer in a room with uni- form absorption. This feature can be used to reduce the computation time by setting each octave band, for ex- ample, to a different diffusion coefficient.

SUMMARY

A multiplicity of analysis tools are available in some room acoustic modeling programs, that allow detailed studies of the effectiveness of altering the surface diffu- sion coefficients. However, there remains a gap be- tween the proposed definition of this parameter for a FIGURE 2. Sound rose depictions for 0% surface diffusion measured physical surface, and its implementation in a (top) and 100% diffusion (bottom) with uniform 15% absorp- modeling program algorithm. It appears that for the tion coefficient on all surfaces. The time window is approxi- program shown here, adequate tools are available to mately the first mean free path after direct arrival. point the way toward closing this gap. Auralization is also a suitable method for observing There are program facilities to alter the source and these differences. Auralized comparisons will be played receiver type, location, directivity, and pointing direc- and unique aural comparative data will be presented. tion. For example, using B-format [4] WXYZ configu- Audio files will be available by contacting the author at ration for the receiver, it is possible to observe impulse the email address given. responses (IR) as one omnidirectional, plus three mutu- ally perpendicular figure-of-eight microphones, as seen in Fig. 3. REFERENCES Receivers can be set to binaural, pointed in a particu- 1. AES Standards Committee, Information Document: lar direction, and the resulting impulse responses can be AES4-ID, www.aes.org auralized and presented on headphones to the best 2. Internet: www.catt.se acoustic analyzer known, the human listener. In addi- 3. Dalenback, B-I, A new model for Room Acoustic Pre- tion, the location of the receiver can be moved about diction and Auralization, PhD Thesis, F95-05, Chalmers the room. A series of locations forming a path can pro- University, Sweden vide a "walk-through" aural experience. 4. Gerzon, M.A., JAES 33(1), 859-71(1985) Comparisons between Computer Simulations of Room Acoustical Parameters and those Measured in Concert Halls Part 2 : Göteborgs Konserthus and Barbican Concert Hall

M. Yugea, H. Shiokawaa, J. H. Rindel b, C. L. Christensen b, A. C. Gade b, M. Itamotoa

aDepartment of Architectural Engineering,College of Industrial Technology, Nihon University, 1-2-1 Izumi-cho Narashino-shi 275-8575 Chiba, Japan bAcoustical Technology, Oersted-DTU, Technical University of Denmark, Build. 352, 2800 Lyngby, Denmark

A number of European concert halls were surveyed in 1989[1]. In this paper comparisons are newly made between measured room acoustical parameters and those obtained from computer simulations using the ODEON program version 4.2 on other two concert halls. One is Konserthus in Göteborg and the other is Barbican concert hall in London. These concert halls are not rectangular box shapes.

GEOMETRICAL DATA FROM COMPUTER SIMULATIONS

Two model of each concert hall are used for computer simulation. One is made with high geometrical fidelity (C.H.) and the other is made with simplifications to geometry, mainly in the platform and reflectors (C.S.). The geometric details of each hall are shown in Table 1. C.H. model of Konserthus in Göteborg and Barbican concert hall in London are shown in Fig.1 and 2. Source and receiver positions in simulation models are defined according to survey in 1989. S1 () is the source position, which is one meter above the floor. R1~5 () are receiver positions in audience area, which are 1.2 meters above the floor. FIGURE 1. C.H. model of Göteborgs Konserthus Acoustical Parameters

Reverberation time (RT): Comparisons of measured (M) and calculated RT on receiver position R2 in Göteborgs Konserthus when the source position is S1 (typical soloist position) are shown in Fig.3. Those of RT on receiver position R1 in Barbican concert hall

Table 1 Geometrical data for each hall Göteborg Barbican C.H. C.S. C.H. C.S. Actual Volume [m3] 11,900 17,750 Estimate Volume [m3] 12,323 12,500 20,975 19,800 Number of seats 1,286 2,026 Number of surfaces 229 91 320 130 C.H.: Computed, High degree of detail FIGURE 2. C.H. model of Barbican Concert Hall C.S. : Computed, Simplified geometry

3.2 3.2 S1-R2 M S1-R1 M 2.8 2.8 C.H. C.H. 2.4 C.S. 2.4 C.S.

2.0 2.0 RT [s] RT RT [s] RT 1.6 1.6 1.2 1.2 125 250 500 1k 2k 4k 125 250 500 1k 2k 4k Octave band center frequency [Hz] Octave band center frequency [Hz] FIGURE 3. RT in Göteborgs Konserthus (S1-R2) FIGURE 4. RT in Barbican Concert Hall (S1-R1)

8.0 8.0 500[Hz] C.H. 500[Hz] C.H. 6.0 C.S. 6.0 C.S.

4.0 4.0

2.0 2.0 Relative error Relative error Relative

0.0 0.0 RT EDT L C Ts LEF RT EDT L C Ts LEF Acoustical parameters Acoustical parameters FIGURE 5. Relative errors at 500Hz in units of FIGURE 6. Relative errors at 500Hz in units of subjective limen in Göteborgs Konserthus subjective limen in Barbican Concert Hall

8.0 8.0 2k[Hz] C.H. 2k[Hz] C.H. 6.0 C.S. 6.0 C.S.

4.0 4.0

Relative error Relative 2.0 2.0 Relative error Relative

0.0 0.0 RT EDT L C Ts RT EDT L C Ts FIGURE 7. RelativeAcoustical errors at parameters 2kHz in units of FIGURE 8. RelativeAcoustical errors at parameters 2kHz in units of subjective limen in Göteborgs Konserthus subjective limen in Barbican Concert Hall are shown in Fig.4. Three series of RT are displayed in Fig.3 and 4; measured RT in 1989 [1] and calculated Conclusion RT by computer simulations using models of C.H. and C.S.. All three series of RT are similar except RT at Measured and calculated reverberation time 250Hz of Fig.4. insingle positions are in good agreement. Accuracy rating of acoustical parameters: The Simplified or detailed platform geometry is average of all resulting relative errors for each notimportant for results in the auditorium. acoustical parameter between M and C.H., C.S. at In both concert halls, the average relative errors 500Hz are shown in Fig.5 (Göteborgs) and Fig.6 of 2kHz are smaller than those of 500Hz. (Barbican). Those of 2kHz are shown Fig.7 and 8 Models with simplifications to the geometry are (Subjective limen: 5% for RT and EDT, 1dB for L and more useful than those with high geometrical C, 10ms for Ts, 0.05 for LF [2]). fidelity in the Barbican concert hall. Relative Errors of EDT at 500Hz are large (3-5 sub. limen) in both concert halls. According to Fig.5, any other Relative Errors are small (1-2 sub. limen) except REFERENCES RT and Ts of C.S.. According to Fig.6, Relative Errors 1. A.C.Gade: Acoustical Survey of Eleven European Concert Hall, of Ts are large (2-4 sub. limen). DTU Report No.44, 1989 Relative Errors at 2kHz are small (1-2 sub. limen) except EDT in Göteborgs Konserthus and Ts of C.S. in 2. Michael Vorlander: International Robin on Room Acoustical Barbican concert hall. Computer Simulation, 15th International Congress of Acoustics, 1995 3. Jens.H.Rindel, Hiroyoshi Shiokawa, Claus.L. Christensen and A.C.Gade: Comparisons between computer simulations of room acoustical parameters and those measured in concert halls, PROCEEDINGS Joint Meeting of the Acoustical Society of America and European Acoustical Association, 1999 Adequacy of the Degree of Cross-correlation between Two Points in Evaluation of Auditory Source Width K. Nakagawaa,b, M. Morimotoa, M. Jinyaa and K. Ohtsua aEnvironmental Acoustiocs Laboratry, Faculty of Engineering, Kobe University, Nada, 657-8501 Kobe, Japan bEnvironmental Engineering Group, Nikken Sekkei Co.Ltd., 541-8528 Osaka, Japan

Auditory source width (ASW) is an important factor for subjective judgement of concert halls. The degree of interaural cross-correlation (ICC) has been proposed as a physical measure for ASW. However a dummy head, which is necessary for measurements of ICC, is expensive and requires troublesome transpotation. In this study, the adequacy of the degree of cross-correlation between two points without a dummy head was investigated. Degrees of cross-correlations measured with two microphones (TMCC) and those with a ball, on which two microphones were installed (SMCC), were compared with ICC using 1/3 octave band noises. The results demonstated a possibility to approximately obtain IACCE3, which is a physical measure for evaluating ASW of music motif, without a dummy head.

INTRODUCTION the head of KEMAR.

Auditory source width (ASW) is an important factor for Experimental Results and Discussions subjective judgement of concert halls. It is well-known that interaural cross-correlation (ICC) is a physical measure for Figure 1 shows the measured SMCC15, ICC, and the evaluating ASW.[1,2] However, ICC is not used in practice difference between them. For 500 and 1000Hz, SMCC15 very often, because a dummy head, which is costly and gives agreed with ICC. The differences between them were less a lot of troubles with transportation, is necessary for than 0.1. For 2000Hz, the correlation coefficient was lower measurements of ICC. than 500 and 1000Hz, but the differences between them did In this study, possibilities of predicting ICC by the two not exceed 0.1 for the frontal and lateral reflections. physical measures were investigated. They are the degrees of Although the differences between SMCC15 and ICC were cross-correlations measured with two microphones (TMCC) about 0.2 for reflections from behind the listener, they may and those with a ball on which two microphones were be negligible. Because early reflections dominantly installed (SMCC). contribute to ASW comparing with late reflections and most of them come from frontal directions. For 4000 and 8000Hz, METHODS SMCC15 differed from ICC very much. SMCC18 agreed with ICC only for 500Hz, but for other The test sound field consisted of a direct sound and two frequencies, the differences between them were larger than lateral reflections in an anechoic chamber. The direct sound 0.2 for some directions of reflections, and the correlation was radiated from the front of the receiving point, and the coefficients between them were less than 0.6. reflections were radiated symmetrically relative to the Accordingly, SMCC15 is predictable of ICC for 500, receiving point. On the receiving point, KEMAR dummy 1000, and 2000Hz. For 4000 and 8000Hz, it is difficult to head or two omni-directional microphones were arranged. obtain the same value of SMCC as ICC because of two The distance between the center of the dummy head or the reasons; one is the difference between directivities of a microphones and the loudspeakers was 1.5m. As source sphere and a dummy head, and the other is the difference sighals, 1/3 octave band noises of fc=500, 1000, 2000, 4000, between the equivalent interaural distance of a dummy head and 8000Hz were used. The direct sound and the reflections and the distance between two microphones. were incoherent each other. The relative sound pressure level of each reflection to a direct sound was fixed of -8dB for 500 CROSS-CORRELATION MEASURED and 1000Hz, and -6dB for 2000, 4000, and 8000Hz. WITH TWO MICROPHONES(TMCC) CROSS-CORRELATION MEASURED Experimental Conditions and Results 1 WITH A BALL AND TWO MICROPHONES (SMCC) The degree of cross-correlation measured with two omni- directional microphones separated by 26, 30, and 33cm Experimental Conditions (TMCC26, -30, and -33, respectively) were compared with ICC. 26cm is the average of equivalent interaural distance of The degrees of cross-correlation measured with a ball of KEMAR for the low frequency region in a free field. 33cm 15cm and 18cm diameter, on which two microphones were is the average of equivalent interaural distance over all installed (SMCC15 and SMCC18), were compared with the frequencies in a diffuse sound field. 30cm was adopted as a degrees of ICC. medium value between 26cm and 33cm. Figure 2 shows that 15cm in length is much the same as head breadth of TMCC30 agreed with ICC for 500Hz. However, for other KEMAR [3], and 18cm is almost same as the distance round frequencies, any TMCC was not in agreement with ICC. For 2000Hz, the difference between TMCC21 and ICC Experimental Conditions and Results 2 was not very small compared with TMCC23 and ICC. However TMCC21 had a higher correlation with ICC than From the relation of the equivalent interaural distance and TMCC23. the wave length, it was expected that the value of TMCC nears to the value of ICC by shortening the distance CONCLUSIONS between two microphones for 1000 and 2000Hz. As a result, 23 and 21cm were adopted for 1000 and 2000Hz, SMCC measured with the ball of 15cm diameter agreed respectively, by comparing the equivalent interaural with ICC for 500, 1000, and 2000Hz, and TMCC measured distances for each direction with the distance between two with two microphones separated by 30cm and 21cm agreed microphones. Figure 3 shows the degrees of cross- with ICC for 500Hz and for 1000 and 2000Hz, respectively. correlation measured with two omni-directional These results indicate a possibility to obtain IACCE3 [2] microphones separated by 21 and 23cm (TMCC21 and -23) approximately without a dummy head, that is, the average and ICC for 1000 and 2000Hz. value of three bands of 500, 1000, and 2000Hz by TMCC For 1000Hz, the difference between TMCC21 and ICC and SMCC. was less than about 0.1 for every reflection, and the difference was smaller than TMCC23 and ICC in lateral. REFERENCES TMCC21 had a higher correlation with ICC than TMCC23. 1. M.Morimoto, H.Setoyama and K.Iida, "Consistent physical measures of auditory source width for various frequency components of reflections", in Proceedings of the Acoust. Soc. Am. and Acoust. Soc. Jpn. Third Joint Meeting, 83 (1996) 2. T.Hidaka, L.L.Beranek and T.Okano, J. Acoust. Soc. Am. 98 (2), 988 (1995) 3. M.D.Burkhard and M.Sachs, J. Acoust. Soc. Am. 58, 214 (1975)

1 1 1 1 500Hz 1000Hz 2000Hz TMCC30 0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4 r=1.000 r=0.979 r=0.625 0.2 0.2 0.2 0.2 r=0.993

0 0 0 Degree ofcorrelation cross 0 Degree ofcorrelation cross 0 36 72 108 144 180 0 36 72 108 144 180 0 36 72 108 144 180 0 36 72 108 144 180 Azimuth angle of reflections (degree) A zimuth angle of reflections (degree)

FIGURE 1. ICC (closed circle), SMCC15 (open circle), and the difference FIGURE 2. ICC (closed circle), between ICC and SMCC15 (open triangle) as a function of azimuth angle TMCC30 (open circle) and the of reflections for 500, 1000, and 2000Hz. difference between ICC and TMCC30 (open triangle) for 500Hz.

1 1 1 1 (a) TMCC21,1000Hz (b) TMCC23,1000Hz (c) TMCC21,2000Hz (d) TMCC23, 2000Hz 0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6

0.4 0.4 0.4 0.4 r=0.672 r=0.563 r=0.990 r=0.900 0.2 0.2 0.2 0.2

Degree ofcorrelation cross 0 0 0 0 0 36 72 108 144 180 0 36 72 108 144 180 0 36 72 108 144 180 0 36 72 108 144 180 Azimuth angle of reflections (degree)

FIGURE 3. ICC (closed circle), TMCC (open circle), and the difference between ICC and TMCC (open triangle) as a function of azimuth angle of reflections. (a) TMCC21, 1000Hz (b) TMCC23, 1000Hz (c) TMCC21, 2000Hz (d) TMCC23, 2000Hz A Method for the Evaluation of the Acoustics of Auditoriums G. P. Guimarãesa, C. G. Herreraa, H. Camille Yehiaa , E. Bauzer Medeirosb aDepartamento de Engenharia Eletrônica, bDepartamento de Engenharia Mecânica Universidade Federal de Minas Gerais, Av. Antônio Carlos 6627, 31270-901, Belo Horizonte, MG, BRAZIL Tel: + 55 31 34995247, Fax: + 55 31 34433783, email: [email protected], [email protected] The present work addresses the problem of auditorium evaluation by means of a sequence of objective measurements. The excitation employed in the experiments has been generated by digital signal processing techniques with the objective of trying to obtain more adequate excitation Maintaining both a moderate cost and a reasonable flexibility for the whole procedure has also been one of chief guidelines of this project, as the authors hope that standardized testing procedures should become more common in the future for small auditoriums. TSP pulse techniques have been employed to evaluate a typical auditorium’s response, using to a larger extent the auditorium’s facilities. Prior to testing, the sound source has also been characterised in quasi non reverberant field conditions. The main results for a typical auditorium are presented together with a discussion of the testing procedure.

INTRODUCTION

The quantification of acoustic properties in auditoriums is possibly one of the most difficult tasks which can be requested from an acoustician. Not only are the actual physical processes involved but also associating objective parameters to the subjective evaluation of a qualified listener, e.g. a musician, can be a challenging exercise. In ancient times some of the attributes were already known [1], but it was only after the pioneering work of Sabine, many centuries later, that the basis for the correlation between objective parameters and subjective evaluation was laid down. Reverberation time represented the predominant thinking in all the early work, with loudness and clarity also appearing sometimes. Recent research such as [1,2], has provided a wealth of detail and objective FIGURE 1: The tested auditorium rating information which should hopefully become the Each pulse was initially computer generated and rule for future analysis. recorded in a CD, providing a consistent excitation Objective rating demands adequate measurement signal for every test condition. The response was techniques. Modern digital signal processing, and recorded in DAT media, providing the means for later adequate instrumentation can provide an efficient tool computer signal processing and analysis. Prior to TSP for objective parameter rating. testing, the auditorium’s reverberation time was verified and found to be 1.4 seconds (average). EXPERIMENT OVERVIEW TESTING PROCEDURE The auditorium of the UFMG School of Music was chosen, for a variety of reasons, as the test object of the Actual testing consisted of the following steps: present analysis. Testing consisted essentially of ! Testing signal generation and CD recording; applying a controlled TSP signal [3,4], for a variety of ! Sound Source (loudspeaker) characterising; conditions, and recording the response. The advantage ! Auditorium testing; of this procedure is to provide relatively quickly an ! Signal processing and analysis. overall evaluation of an auditorium’s response. Prior to any auditorium testing the sound source, one Figure 1 shows a simplified diagram of the chosen of the auditorium’s loudspeakers had to be auditorium for a typical test condition, also indicating characterised. The sound source and all the testing the source and the microphone relative position. equipment was taken to a large grass field inside the university campus, and measurements were carried out causes related to FFT evaluation, and is presently in the early hours of the morning, in order to obtain a under the object of a current research by the authors. low level of background noise and reverberation. The measurement distances were the same used inside the auditorium. The background noise level was measured and subtracted from the final result. As in the auditorium’s case, calibration was based on a value of 85.5 dB SPL, corresponding to – 3 dB on the DAT recording system. For the auditorium a similar procedure has been employed. Measurements were carried out for several test signals and geometric configurations. Figure 1 indicates the reference conditions, with the loudspeaker at the same height as the microphone (1,80 m), the horizontal distance between them being 5 metres.

RESULTS AND CONCLUSIONS FIGURE 3: Auditorium plus Speaker Response Figures 2, 3 and 4 show, respectively, the speaker open field response, the combined speaker and auditorium response, and the corrected auditorium’s 1/3 octave frequency response, the standard deviation, the vertical lines, being also shown for every case.

FIGURE 4: Auditorium Corrected Response

Finally, it should be mentioned that the present research is being now extended, in order to associate standard evaluation parameters with the present results FIGURE 2: Loudspeaker Response and with the subjective (listening) evaluation of the Escola de Música musicians. The results correspond to a TSP of 0.34 seconds, ranging from 20 Hz to 20 kHz. The recorded files have ACKNOWLEDGEMENTS been “cut off” at 0.85 seconds (open field) and 1.0 The authors would like to thank Escola de Música da seconds ( auditorium ), to reduce the influence of the UFMG for the kindly offered support. They would also background noise level on the experiment. like to acknowledge the financial support provided by The signals have been recorded in .WAV extension, FAPEMIG which has enabled the acquisition of part of subsequently, FFT treated. The combined response of the equipment employed in the present research. the auditorium may be found by comparison of the speaker (open field ) response, with the response measured at the auditorium. The results here shown REFERENCES correspond to an average evaluated for five 1. L. L. Beranek, Concert and Opera Halls: How they consecutive recordings, of the same TSP, observing the sound, ASA, (1996). attenuation time for each pulse. 2. T. Hidaka and L. L. Beranek, Journal of the Acoustic It is interesting to observe that the standard Society of America, 107, 368-383, (2000). deviation has a larger value at the lower frequencies of 3. Y Suzuki et al., Journal of the Acoustic Society of the spectrum. This is likely to be associated with America, 97, 1119-1123, (1995). 4. N. Aoshima, Journal of the Acoustic Society of America, 69, 1484-1488, (1981). A Study on the Acoustic Mechanisms for Supplying Reflection from Behind without the Use of Back Wall Effects K.Sekiguchia, T. Hanyua and Y.Murohashic

aCollege of Science and Technology,Nihon University,1-8 Kanda Surugadai Chiyoda_ku Tokyo,1018308,Japan bWakabayashi onkyo limt.,3-69-1Hatagaya Shibuya_ku Tokyo, 151-0072 Japan

Listener Envelopment (LEV) becomes larger when a reflection arrives and envelops the listeners [1]. In order to do this, we have to focus on and to figure out how we can supply a reflection from behind, which should be in good proportion to the initial reflection that frequently arrives from the front. In fact, when going through the results of the analysis of the reflection arrival direction in a shoe box with high evaluation results, we can see a lot of reflection arriving from the quarter back area of side walls during the period of time when a back wall effect is not involved. Therefore, side wall designs are the focal point of this study, and we report the results after examining the acoustic mechanisms, which supply reflection from behind (which is a relatively important factor for LEV[2]) to the sound receiving point without using the back wall. After having examined the designs of the side walls in a concert hall, the results clarify that a column type diffuser can create a specific mechanism which supplies and envelop listeners even when no back wall effects are involved.

20cm Y Z 72cm INTRODUCTION

S Side wall designs are the focal point of this study, and we report the results after examining the acoustic 35cm 24cm 3cm R S mechanisms, which supply reflection from behind to the R 72cm X Y sound receiving point without using the back wall. Cross Sectional断面図 View 10cm S: Sound Resources R: Sound Receiving Point EXPERIMENTAL METHOD Plan Side側壁模型 wall model 平面図 1/50 scale models were used for this experiment. In FIGURE1. The arrangement plan of the 1/50 scale model for order to examine the first reflection from the side wall, we this experiment created models that have only side walls, using a shoe 0.5m box type of structure at the size of 20m × 36m × 17.5m, as shown in Figure 1. Three different measurement patterns 0.5m (Patterns , and ) were set in order to understand the Front Sound effects of the reflections from behind. As shown in Diffusion Absorption Figure 2, Pattern is set to diffuse on the entire side wall, Pattern is set to diffuse on the front half of the side wall 2.5m and to absorb sound on the other half, and Pattern is The Whole to absorb sound on the front half and diffuse on the Diffusion other half. Regarding the side wall types, we used three Sound Back 0.5m different types, which are shown in Figure 3. One is a Absorption Diffusion column type side wall, where columns (the actual size is set to be 50 cm in diameter) were placed at intervals of 2.5m 2.5cm, which attach to the side wall, (each attaching place Column Waved Board is numbered from one to thirteen). Another types are a Rimpled Board waved boards, which were waved at the same intervals FIGURE2. Measuring Patterns FIGURE3. Side Wall Types as that of the column type. Styrene board (5mm in thickness), which is considered to have almost perfect reflection, was used as the material for the 1/50 scale pulse, which can be applicable up to 160kHz, is used for models, and plastic was also used for the columns of 1 cm the sound resources and a 1/8 inch microphone was used diameter, which is 50 cm of the actual size. A discharge for receiving the sounds. EXPERIMENTAL RESULTS AND FUR- type ) contributes to a certain sound receiving point is limited, when the above result is examined from the THER CONSIDERATION viewpoint of the spatial distribution. On the other hand, the column type supplies reflections not only from the Figure 4 shows the impulse responses in all three columns on the front side of the wall but also from the different types of side walls in accordance with the three columns placed on the quarter back area where sound different measurement patterns. Figure 4 notes the reflection goes behind the listeners as time passes. periods of arrival time regarding how long it takes for the As a result, this enables reflections to arrive and envelop reflections to arrive from each diffuser (all the diffusers the listeners. have been numbered, as seen in Figure 2). When considering these factors and comparing the waves in Figure 4, it is understood from Pattern that the CONCLUSION responses in both the rimpled and waved types are seen during the initial stage of response, while the responses After having examined the designs of the side walls in a of the column type is distributed over a wide range of concert hall, each column produces the cylindrical wave time. When looking at the waves of Pattern , the on the side wall of which the column has been placed in reflections in all the wall types are seen focused upon during the initial stage. Pattern shows the same result the lined pillar state. Regarding the side wall where as Pattern , showing that the responses of both the columns were placed in a line, each column can produce rimpled and the waved types of side walls are mainly spherical waves and cylindrical waves. As a result, it can seen as focused during the first stage of responses, while be assumed to create an acoustical mechanism which can the reflection in the column type is seen as distributed supply sound reflections from various different directions over a wide range of time, which makes it clear to to listeners in wider areas during the initial stage. understand that the reflection arrives from the columns located in the back. These factors are assumed to have great effects on the function of enlarging the amount of REFERENCES Ts, which the authors have been using for determining the amount of LEV. [1] T. Hanyu and S. Kimura., Applied Acoustics 62(2), The part which the first reflected sound from diffusing 155-184(2001). wall surface ( the rimpled board type and the waved board [2] M.Morimoto and K.Iida, J.Acoust.Soc.Am.98, 2282 (1993)

Column I Rimpled Board I Waved Board I

5,6 3 2 1 5,6 3 2 1 5,6 3 2 1 4137 8 9 10 11 12 4137 8 91011 12 4137 8 91011 12

Column II Rimpled Board II Waved Board II

4 4 4 5,6 7 3 2 1 5,67 3 2 1 5,6 7 3 2 1 Column III Rimpled Board III Waved Board III

7138 910 11 12 7138 91011 12 7138 91011 12 FIGURE4. Impulse responses of all the three side wall types (120ms)

Contrast of Spatial Impression due to the Variation of Musical Notes in a Concert Hall

T. Hanyua, K. Sekiguchib and Y. Satakec

aDepartment of Construction, Nihon University Junior College, 7-24-1 Narashinodai, Funabashi, Chiba, Japan bDepartment of Architecture, College of Sci. & Tech., Nihon University, 1-8 Surugadai, Kanda, Tokyo, Japan cNittobo Acoustic Engineering Co.,Ltd., 1-19-9 Midori, Sumidaku, Tokyo, Japan

While we are actually listening to music in a concert hall, we sometimes experience changes of spatial impressions such as Listener Envelopment (LEV), as music is playing. In this study, we conducted psychological experiments in order to investigate the phenomenon of the changes of LEV due to the variation of musical notes. As a result, we clarified that LEV is occasionally changed due to the compositions of sound from a variety of instruments, tunes and tempos even in the same sound field. Furthermore it was clarified that each hall has characteristics of change of LEV. If change of LEV is large, a listener felt the contrast of LEV. It was also understood that contrast of LEV was occurred in the shoebox type hall that generally has a good acoustic condition. In the fan shape hall its LEV was always small regardless of the variation of notes. On the other hand, the LEV was always large in the diffuse sound field (such as in church). Therefore, in these sound fields no contrast of LEV occurred by the variation of notes.

INTRODUCTION Verdi, were chosen to be used as music for our tests since this piece is enriched with various changes in its It is considered that the spatial impression in a concert composition such as the use of different instruments, hall is occasionally changed due to the compositions of varied tempos and duration of notes. The level of sound from a variety of instruments, tunes and tempos. stimuli at the listening point was 75dBA. Therefore, these changes of spatial impressions can be used as a new measure if they can be confirmed to be First, LEV in each sound field was obtained in Test A. different in each different hall. In this study, degrees The test was conducted according to Scheffe’s paired of changes in LEV, which can be dynamically changed comparison method. Each pair of sound fields was as music is playing, were examined in various sound randomly presented. Seven subjects rated the LEV of fields. the second sound field of each pair on a seven-point scale in comparison with the first one.

TEST METHOD Next, two tests were conducted according to the SD method in order to examine “LEV Contrast. ” Seven A total of three tests were conducted using a simulated subjects rated the “LEV Contrast” of each sound field sound field in an anechoic chamber. We installed 16 on a seven-point scale from “no contrast” to “very loudspeakers at a distance of 1.5m from the listeners at contrasted.” In addition, according to “LEV Contrast,” equal intervals of 22.5 degrees in the horizontal plane. focusing on differences in musical instruments (Test Six different types of sound fields were used in these B) and in ways of playing (Test C), the subjects were tests. Figure 1 shows the plan and the virtual sound told to say there was a contrast whenever they felt source distributions (VSSD) of six sound fields, as major differences in each LEV. In Test B, in the parts well as the responses synthesized from VSSD. The (Measures 29 ~ 37) composed of the main melody of VSSD of each hall was measured in actual concert cellos and the high notes of violins, differences in LEV halls by using 4-point microphone system [1]. between these musical instruments were evaluated Artificial sound fields E and F are formed from focusing on each instrument. In Test C, differences in randomly generated virtual sound sources so that LEV were evaluated focusing on portions where reflected sound is diffused. The directional responses violins were played slowly for long durations required for reproduction were synthesized from the (Measures 18 ~ 27) and also focusing on parts where VSSD. As a stimulus, we used a sound composed of violins were played fast for short durations (Measures the directional response waveform convolved with an 36 ~ 41). The subjects repeatedly listened to stimuli anechoic music. Measures 17 ~ 41 of the “Overture to until they were satisfied with their evaluation. the First Act of La Traviata,” an opera composed by

characteristics and sound field characteristics. Since 500500msms 500500msms 500500msms there are the differences concerning degrees of LEV front front front Contrast in each different sound field, LEV Contrast can be confirmed as an element of evaluation of sound left right left right left right field. We conjecture that the sound field where LEV Contrast occurs might be more sensational. In fact,

rear 200ｍｓ rear rear LEV Contrast was occurred in the shoebox type hall

that generally has a good acoustic condition. On the

16 １ 2 15 3 16 １ 2 15 3 14 4 other hand, in the fan shape hall its LEV was always RT:3.2s 14 4 RT:2.3s １ RT:1.7s 13 16 2 5 13 5 15 3 12 6 12 6 14 4 13 5 small and in the diffuse sound fields a listener was 11 7 11 7 10 8 12 6 10 9 8 9 11 7 10 9 8 always surrounded by sound. Therefore, in these Hall A Hall B Hall C sound fields no contrast of LEV occurred. We suspect that LEV Contrast is related to the acclimation of 500500msms 500500msms 500500msms binaural hearing [3]. It seems that the sensitivity of front front front the spatial impression becomes dull due to this acclimation. Hafter et al. suggested that a certain left right left right left right trigger is needed to cancel the acclimation [3]. It is thought that this acclimation might be canceled by the

rear 200ｍｓ rear 200ｍｓ rear changes of spatial impression such as LEV Contrast. Artificial sound field: (which is formed from

16 １ 2 randomly generated virtual sound sources 15 3 when we assumed reflected sound is diffused) 3 120 14 4 13 5 Y(0.05)=0.41 100 12 6 2 11 7 RT:2.1s RT:3.0s RT:3.0s 10 9 8 80 Hall D Sound Field E Sound Field F 1 0 60 FIGURE 1. Sound fields used in tests. -1 SBTs[ms] 40 -2 20 RESULTS AND DISCUSSION -3 0 Psychological Scale of LEV A B C D E F A B C D E F Sound Fields Sound Fields Figure 2 shows both the psychological scale of LEV in Test A and the spatially balanced center time (SBTs) FIGURE 2. Psychological scale of LEV in Test A (proposed by the authors as an evaluation index of (left) and SBTs in each sound field (right). LEV) [2] obtained in each sound field. Shoebox type hall A and diffused sound fields E and F especially 76 76 show high LEV. The values of SBTs are seen to Test B Test C 65 65 correspond to psychological scores of LEV, and the 54 54 correlation coefficient between them was confirmed to 43 43 be 0.98 (p<0.01). 32 32 Contrast of LEV 21 Contrast of LEV 21 Psychological scales in Tests B and C are shown in 10 10 A B C D E F A B C D E F Figure 3. The error bars in this figure show 95% of the confidence interval. LEV Contrasts shown in Sound Fields Sound Fields Figure 3 are not always seen to correspond to the LEV FIGURE 3. Psychological scales of Tests B and C. scores shown in Figure 2. Although the scores of LEV in sound fields E and F are high as shown in Figure 2, LEV Contrasts are low scores as shown in Figure 3. Therefore, it can be understood that sound fields E and REFERENCES F always have a high LEV without any influence from different kinds of musical instruments or ways of 1. K. Sekiguchi, S. Kimura, and T. Hanyu., Applied playing them. On the contrary, as hall A shows high Acoustics 37, 305-323 (1992). scores in both “LEV” and “LEV Contrast,” it is assumed that this can be the sound field where LEV is 2. T. Hanyu and S. Kimura., Applied Acoustics 62(2), 155- changed due to the variety of musical instruments, 184(2001). tempos and duration of notes. As sound field is not changed, it is considered that this phenomenon can be 3. E. R. Hafter, R. H. Dye and E. M. Wenzel., J. Acoust. Soc. Am. 73, 1708-1713(1983). caused by the relationship between sound