<<

Neutralizing the meaning of sound for sound evaluations

H. Fastl

Institute for Man-Machine-, Technical University München, Arcisstr. 21, 80333 München, Germany

Sound quality usually can be described by psychoacoustic magnitudes like , sharpness or roughness. However, sometimes the meaning of a sound may strongly influence its subjective evaluation. In order to assess these effects, sounds „without meaning“ have to be realized. A frequently applied solution of this problem is to fill the temporal envelope of an original sound with pink . However, this procedure has the severe disadvantage that the original sound and the sound with no meaning differ in loudness. Therefore, a procedure is proposed, which can „neutralize“ the meaning of sound, and at the same time preserve the original loudness.

INTRODUCTION These differences are of great relevance, since the evaluation of sound quality in many cases crucially For the assessment of sound quality, psychoacoustic depends on the loudness of the sound ([1], [5]). magnitudes like loudness, sharpness, fluctuation strength or roughness have proven successful. In many cases, a combination of these basic psychoacoustic magnitudes can predict sound quality ratings by subjects [1]. Despite the fact that the application of these principles usually leads to successful engineering results, sometimes the meaning of a sound may considerably influence its evaluation. For example it could be verified both in the field (see [2]) and in the laboratory (see [3]) that at same equivalent A-weighted level, railway noise is preferred to road traffic noise. From an engineering approach it could be shown [4] that spectral differences between railway noise and road traffic noise are reflected in the FIGURE 1. Traditional procedure to remove meaning of specific loudness patterns and can account partly for sound. the advantage of the railway noise. Also the time- The loudness differences to be expected because of structure of noise immissions from railway noise and different bandwidth are illustrated in figure 2. The left road traffic noise differ significantly. In addition to part of figure 2 shows the dependence of the loudness these differences, which can be assessed by physical of (PN) or an 1 kHz-tone on level. At same means, there may be also an influence of the different level, the loudness of the pink noise is much larger meanings of the sounds [3]. than the loudness of the . For example at 94 dB, as indicated by the vertical double arrow, the PROCEDURES loudness of the pink noise is about a factor of 2.5 larger than the loudness of the 1 kHz-tone. In order to neutralize the meaning of sound, in several The right part of figure 2 shows original sounds (left labs the following procedure is used (cf. Figure 1). column) and sounds with the same temporal envelope, From the original sound, the temporal envelope is filled with pink noise (right column). A comparison of extracted, e.g. by rectifying an low-pass filtering. In a the left and right column reveals that the temporal modulator, this is multiplied with pink noise, structure of original and synthesized sounds is very leading to a sound which has the same temporal similar. However, in many cases the loudness of the envelope as the original sound but no meaning. synthesized sounds is larger. This discrepancy is However, because loudness depends on the bandwidth illustrated in the left part of figure 2 by vertical lines. of a sound, even if the temporal envelope is faithfully For example sound (a), when replaced by pink noise extracted and filled with pink noise, the original sound with same temporal structure (b), increases from and the synthesized sound usually differ in loudness. around 40 to around 60 sone. Likewise, the simulations (d) and (f) of the sounds (c) and (e) show a original sounds, but also preserves their loudness in loudness which is approximately 50% larger than the great detail. loudness of the original sounds.

FIGURE 2. Loudness differences of sounds with same level but different bandwidth. FIGURE 4. Comparison of original sound (a) and corres- ponding sound without meaning (b). Because loudness is such a dominant cue in sound quality evaluation, we looked for possibilities to OUTLOOK preserve the temporal envelope and also the loudness. In this respect, the Fourier-Time-Transform (FTT [6], With the tool illustrated in figure 3, we now have the [7]) proved to be a very helpful tool. The principle of possibility to study sounds which are as much as the procedure applied is outlined in figure 3. The possible identical with the crucial difference that one is original sound first is analysed by the FTT-procedure. an original sound with a specific meaning, and the After spectral broadening of the elements of the FTT- other is the corresponding synthesized sound, from patterns, sounds are sythesized by an inverse FTT which the meaning was removed. algorithm. As a result we get a synthesized sound, which has the same temporal envelope, no meaning ACKNOWLEDGEMENTS and – in contrast to the procedure illustrated in figure 1 – also the same loudness-time function as the original The author wishes to thank the members of his group sound. “Technical ” for support in realizing the sounds and editorial help.

REFERENCES

1. Zwicker, E., Fastl, H., . Facts and Models. 2nd updated ed., Springer-Verlag, Berlin, 1999. 2. Möhler, U., Community response to railway noise: a re- view of social serveys, J. Sound Vib. 120, 321-331, 1988. 3. Fastl, H., Kuwano, S., Namba, S., Assessing in the railway bonus in laboratory studies. J. Acoust. Soc. Jpn. (E) 17, 139-148, 1996. 4. Fastl, H., Masking effects and loudness evaluation. In: Recent Trends in Research (H. Fastl et al. Eds.) Bibliotheks- und Informationssystem der Carl von FIGURE 3. New procedure to remove meaning of sound. Ossietzky Universität Oldenburg, Oldenburg, 29-50, 1996. Figure 4 shows for comparison original sounds and 5. Fastl, H., Sound Quality of Electric Razors - Effects of their counterparts with neutral meaning obtained by the Loudness. In: Proc. inter-noise'2000, CD-ROM, 2000. procedure illustrated in figure 3. When comparing 6. Terhardt, E., Fourier transformation of time : Con- ceptual revision. Acustica 57, 242-256, 1985. loudness-time functions displayed in figure 4a vs. 7. Mummert, M., Sprachcodierung durch Konturierung figure 4b, there are almost no differences discernible. eines gehörangepaßten Spektogramms und ihre Anwen- This means that the proposed FTT-based procedure not dung zur Datenreduktion. VDI Reihe 10, Nr. 522, VDI only faithfully reproduces the temporal envelope of the Verlag, Düsseldorf, 1998. Dimensions of Sound Quality and Their Measurement

S. Kuwanoa and S. Nambab

a Osaka University, Japan b Takarazuka University of Art and Design, Japan

It is expected to improve sound quality of machinery noise as well as to reduce the sound level. It is important to find physical metrics which show good correlation with subjective impression in order to predict the sound quality and find appropriate countermeasures. In this paper, the validity of physical metrics of sound quality will be discussed on the basis of the results of psychological experiments, especially focusing attention on the temporal aspects of sounds.

INTRODUCTION components. When the sound contains high frequency components, it causes the impression Much effort has been made to reduce the sound level ‘sharp’ and ‘metallic’. In this case, the impression of of machinery noise. However, since there is a limit to metallic factor can be evaluated by calculated reduce the sound level and machinery give us sharpness [1, 5]. Pleasant factor is related to cognitive information concerning the situation of machines, and cultural factors as well as physical properties of recently it is expected to improve sound quality of sounds. It is difficult to predict pleasantness of sounds machinery noise as well as to reduce the sound level. by physical properties alone. However, in a limited It is important to find physical metrics which show situation, it may be possible to find physical properties good correlation with subjective impression in order to which shows good correlation with subjective predict the sound quality and find countermeasures to impression. For example, equal pleasantness contour improve sound quality. Many physical metrics have for air-conditioner noise was proposed based on LLz been proposed [1]. Some of them usually show good and calculated sharpness [6]. correlation with subjective impression, but others do Hearing is a sensation which conveys information not always. Validity of physical metrics of sound along temporal stream. Therefore, temporal factors quality will be discussed on the basis of the results of have an important effect on hearing. The temporal psychological experiments, especially focusing pattern and temporal condition of sounds were found attention on the temporal aspects of sounds. to have a significant effect on each dimension of sound quality in our former studies. DIMENSIONS OF SOUND QUALITY (1) Temporal condition and powerful Sound quality is multi-dimensional. However in most factor of our former studies [2], three main factors, powerful, metallic and pleasant, have consistently been extracted The loudness is usually evaluated by LAeq or LLz as the and they can be regarded as representative factors of first approximation. However, when the temporal sound quality. An example of the result of factor pattern of the sound is systematically varied, the sound analysis is shown in Table 1 [3]. which has high level portion at the beginning is perceived as being louder [7]. This may be due to the RELATION BETWEEN PHYSICAL overshoot at the onset of the sound. METRICS AND DIMENSIONS OF (2) Temporal condition and sharp factor SOUND QUALITY The impression varies according to the duration of Coefficients of correlation between physical metrics sounds [8]. It was found that the shorter the duration and adjective scale values are shown in the right became, the more sharp the sound was judged as columns in Table 1. Good correlation is usually found shown in Fig.1. between LLz (Zwicker’s loudness level based on ISO 532B averaging temporal fluctuation on energy basis) and the impression of powerful factor [4]. LAeq also (3) Temporal condition and pleasant factor shows good correlation with the impression of powerful factor when sounds have broad band An example of the effect of temporal pattern on pleasantness is shown in Fig.2. Temporal changes of significant effect on the impression of sounds even if calculated sharpness of the sounds when a golf ball other physical properties are equal. The temporal was hit by a golf club were found to be different. The effect may be related to the dynamic characteristics of relation between subjective impression of refresh and hearing. It is important to find appropriate method to the temporal change of sharpness is shown in Fig.2. evaluate sound quality taking temporal factors as well High correlation between them suggests that the as other physical properties of sounds into temporal change has a significant effect on subjective consideration. impression. When pulse train is judged, the interval between pulses has a significant effect on the REFERENCES impression. A phrase of a performance (Pictures at an Exhibition composed by Musorgsky) 1. E. Zwicker and H. Fastl, Psychoacoustics,, (Springer, 1999). was played with synthetically varying the interval 2. S.Namba, Measurement of and its Applications, between sounds. It was found that the impression (Oyogijutu Shuppan, 1992). changes systematically with the interval between 3. S. Namba, et al., J.A.S.J. (E), 13, 49-58 (1992). sounds [9]. Similar results were found with gear noise 4. S. Kuwano, et al., Noise Cont. Eng. J., 33, 107-115 (1989). 5. G. von Bismarck, Acustica, 30, 159-172 (1984). in a car [10]. 6. Y. Kikuchi, et al. Proc. of Autumn Meeting of ASJ, 699-700 (1992). 7. S. Namba et al., Jpn. Psychol. Res., 18, 63-72 (1976). FINAL REMARKS 8. S. Namba et al., J.A.S.J., 30, 144-150 (1974). 9. S. Namba et al., Studies of Humanities and Social Sciences, Osaka There are various factors which contribute to the sound University, 41, 17-35 (1993). 10. T. Abe, Dissertation, Osaka University (1995). quality. Temporal factor is one of them and it has a

Table 1 Result of factor analysis

Adjectives Factor 1 Factor 2 Factor 3 r (LLz) r (sharpness) gentle – soft .725 -.114 -.224 .118 .796 distinct – dull -.615 .425 .229 .120 -.722 noisy– quiet -.342 -.387 .546 -.707 -.742 deep – metallic .878 -.036 -.051 .041 .831 pure – impure -.371 .660 -.040 .362 -.571 loud – soft -.139 -.148 -684 -.819 -.590 calm – shrill .842 .044 -.160 .086 .871 – discordant .553 .445 -.060 .321 .838 pleasant – unpleasant .374 .528 -.204 .712 .737 powerful – weak -.073 -.054 .667 -.834 -.582 flat – rumbling .016 .611 -.223 .549 -.200 smooth - harsh .788 .186 -.161 .291 .892

Fig.1

Fig.2 Semantic Attributes of Environmental Sounds and Their Correlations with Psychoacoustic Magnitudes

A. Zeitler and J. Hellbrück

Catholic University of Eichstätt, Germany

Nowadays, elementary psychoaoustical magnitudes such as loudness, sharpness, or roughness can be easily calculated by use of signal analysis software. As sound quality comprises more than elementary sensory attributes, however, additional assessment is needed to get the whole picture of sound quality. In the present study, the semantic differential technique is used to explore both connotative and denotative meanings of a series of environmental sounds. Two independent samples of subjects applied a noise- specific semantic differential to 17 short pieces of noises stemming from sources such as musical instruments, natural environ- ment, technical appliances etc. As a result, test-retest-reliability of the semantic profiles amounted to r=.95. Factor analysis of the 20 adjective scales used in the semantic differential revealed four components, which were interpreted in terms of “evaluation”, “timbre”, “power”, and “temporal change”. Moreover, loudness, sharpness, and roughness were calculated from the signals, and high correlations with respective scales from the semantic differential were found. All in all, results recommend the semantic differential technique as an instrument to explore various aspectes of sound quality with high accuracy.

INTRODUCTION EXPERIMENTS

Various methods are at hand for sound quality as- Method sessment, and at the present time the acoustic engineer is in the position to employ modern signal analysis Subjects tools for the calculation of elementary psychoaoustic magnitudes such as loudness, sharpness, or roughness. A total of 21 subjects (10 male, 11 female) participated However, since the of sounds is dependent in the first experiment. The sample mainly consisted of on cognitive and emotional factors as well, additional students 20 to 56 years of age (median: 26). An inde- measurements are needed to get the whole picture of pendent second sample of 21 students (5 male, 16 sound quality. female), ranging from 19 to 31 years (median: 25), was For decades the use of the semantic differential recruited for a retest six month later. All subjects re- technique, which Osgood [1] developed to identifiy ported normal hearing. emotional meanings of words, has been extended to a large variety of different concepts, including sounds Stimuli [2]. A semantic differential (SD) comprises several pairs of opposite adjectives which constitute the poles Stimuli were 17 environmental sounds which stem- of mostly 7-point bipolar rating scales (see figure 1). med from a broad range of sources such as musical An important methodological issue is that a disctinc- instruments, natural environment, household appli- tion has to be made between connotative and denota- ances, and power tools. The series contained both sta- tive scales. In the present study, the latter refers to tionary and level fluctuating noises (duration: about 5 acoustic or psychoacoustic properties of the sounds seconds). Levels ranged from 61 to 84 db(A) Leq, such as loudness. By contrast, connotative scales are which were intended to reflect natural loudness ratios. intended to measure the emotional meaning contained in the sound on scales such as “calming – exciting”. Procedure In the present study, an SD comprising both types of scales is applied to a series of every-day noises in The task required the subjects to judge the sounds, order to treat the following issues: which were randomly presented through , · Test-retest reliability of the method with a set of 20 bipolar rating scales as depicted in · Factors of sound quality (underlying the SD) figure 1. The presentation of each sound was looped · Correlation of denotative scales with psychoacoustic until all scales had been completed. Prior to the judge- calculations (signal analysis) ments, subjects were presented with all sounds for orientation. Results and Discussion Correlations with Psychoacoustic Calculations

Reliability For each sound, average loudness (Zwicker), sharp- ness (Aures), and roughness were calculated by use of For both groups of subjects (test and retest), individual commercially available signal analysis software (Head judgements were averaged over the adjective scales. acoustics Artemis). The following coefficents result for Thus, for each sound two mean profiles result, as ex- the correlations with the respective adjective scales: emplified in figure 1. Test-retest reliability of the 17 · roughness vs. “smooth-rough”: r=.84 (p<.01) sounds ranges from r=.79 to r=.97, and amounts to · sharpness vs. “dull-sharp”: r=.71 (p<.01) r=.95 for the whole stimulus set. Since no substantial · loudness vs. “soft-loud”: r=.71 (p<.01) differences between the two groups resulted, both In addition, loudness calculations were correlated datasets were aggregated for further analysis. with judgements of a previous study [3], in which the same sounds had been judged using category subdivi- 3 2 1 0 1 2 3 sion (CS) scale. The CS-scale comprises five verbally unpleasant pleasant distinguished categories (“very soft”, “soft”, “me- flat rumbling dium”, “loud”, “very loud”), with a 10-step fine muffled shrill graduation in each category. The correlation between dark light the mean values on this so-called 50-points scale and peaceful aggressive loudness (Zwicker) was r=.91. This correlation dif- ugly beautiful fered significantly from that of the 7-point scale sad bright (p<.05) with loudness. low high soft loud light heavy CONCLUSIONS calming agitating smooth rough Test-retest reliability (r=.95) of mean semantic pro- pure impure files indicates high accuracy of the semantic differen- gentle harsh tial technique. The factorial investigation of the dull sharp slow fast 20 adjective scales revealed four orthogonal factors weak strong which were interpreted in terms of “evaluation”, “tim- boring exciting bre”, “power”, and “temporal change”. unsteady steady Denotative scales of the SD showed high correla- soft hard tions with respective psychoacoustic calculations (loudness, sharpness, roughness), as expected. Com- 3 2 1 0 1 2 3 pared to the 7-point adjective scale, a significantly FIGURE 1: Mean profiles (test and retest ) for stimulus higher correlation for loudness was found with the 50- “hair dryer” (retest-reliability: r=.92). Standard deviations points category subdivision scale. (all noises) ranged from 0.15 to 2.21 scale units.

Factor Analysis ACKNOWLEDGEMENTS

The authors are grateful to Andrea Hahn, Petra Varimax rotated principal component analysis was Schüller, and Mirjam Wolf for their contributions in employed to extract orthogonal factors underlying the this study. 20 adjective scales. With a criterion of eigenvalues > 1, four factors arrived which cover 70 % of the total vari- ance. Variables with high loadings (r>.50) are assigned REFERENCES to the factors as follows: · Factor 1 “Evaluation” (29 %): ugly-beautiful, un- [1] Osgood, C. E., Suci, G. J., and Tannenbaum, P. H. The pleasant-pleasant, calming-agitating, boring- measurement of meaning. University Press of Illinois. exciting, gentle-harsh, pure-impure, soft-hard Urbana 1957. · Factor 2 “Timbre” (17 %): dark-light, low-high, [2] Schick, A. Zeitschrift für Lärmbekämpfung 41(3), 61-68 muffled-shrill, dull-sharp, light-heavy (1994). [3] Zeitler, A. and Hellbrück, J. Psychophysical scaling of · Factor 3 “Power” (16 %): weak-strong, soft-loud, the pleasantness of environmental sounds, in Proceedings flat-rumbling of the 7th Intl. Congress on Sound and . Gar- · Factor 4 “Temporal change” (8 %): unsteady- misch-Partenkirchen, 2000, pp. 2485-2490. steady, smooth-rough Moderators of Sound Quality of complex sounds with multiple tonal components

Markus Bodden*, Ralf Heinrichs**

* Ingenieurbuero Dr. Bodden, Ursulastr. 21, D-45131 Essen, Germany, email: [email protected] ** Ford Werke AG, Acoustic Centre Cologne, Spessartstraße, D-50725 Köln, Germany, email: [email protected]

Tonal components play an important role in the context of Sound Quality for interior vehicle noise. In contrast to basic psycho- acoustic data which are based on experiments with single components embedded in a broadband masker, vehicle sounds are com- posed of a variety of tonal components. The component to be investigated and the do not form clearly different sound sensations any more, and we observed that different subjects focused their attention towards different tonal components while rating the Sound Quality. The results derived from experiments with single tonal components can thus not be transferred to the case of multiple tonal components, and special attention has to be paid to the method to evaluate Sound Quality of these signals.

INTERIOR VEHICLE SOUND class are tonal components, which are caused by va- rious vehicle components: The evaluation of the Sound Quality of interior ve- · the engine sound consists of multiples of the engi- hicle sounds is a complex task. First, the sound is com- ne order posed of a variety of different components, and second, · road noise might include tonal components from the perception of Sound Quality by humans is a com- tires or road surfaces plex process. This process is not only based on the pure · the transmission often creates tonal components physical signal, but also on other modalities like visual · automatic steering, oil and fuel pumps, chains, and or tactile information and even non-sensory moderators all kind of electric motors cause further contributi- (e.g., [1]). ons. In general, a product sound consist of two different Fig. 1 shows that an interior vehicle sound is compo- groups of features, undesired and desired features. The sed of multiple tonal components covering a broad fre- first group comprises the sound features which have to quency range. It is important to note that the be avoided (e.g., squeak and rattle), the second group components often do not show ideal sinusoidal charac- comprises the sound features which form a „good“ ter, but can be much broader. sound (e.g., pleasant, sporty, noble). The undesired features play an important role, be- TONAL COMPONENTS cause they can significantly degrade a good sound and spoil the effort spent to create it. They have to be redu- It is obvious that the evaluation of the influence of ced to that level where they do not degrade the overall the various tonal components is a difficult task, althou- sound (acceptance level). gh the investigations on the perception of tonal com- But, this task is complicated by the fact that the clas- ponents have a long tradition. Basic psychoacoustic sification of specific components might change. Today research data is available from literature with respect to for example the characteristic sound of a turbo charger detection and masking (e.g., [2]), but they basically all is not classified as undesired, because it gives the feed- consider either the case of a single tonal component in a back that the car is equipped with this non-standard fea- broadband, non-tonal masker, or the relation of two to- ture. But in the future, when nearly all vehicles with a nal components without any masker. The description diesel engine will be equipped with a turbo charger, this above shows that these data can thus not directly be ap- feedback information will loose its importance, and the plied to interior vehicle sounds. sound might be classified as undesired. Furthermore, in the context of Sound Quality not the Interior vehicle sound is composed of the major detection of tonal components is important, but their ac- components from engine, road, and wind, plus contri- ceptance level. We thus planned several experiments to butions from various components. Since the overall le- investigate the frequency-dependency, the role of the vel and undesired sounds like squeak and rattle have width of a component, and the influence of multiple to- been significantly reduced in the past, now other sound nal components. components turn out to be undesired. One important All experiments have been conducted in a sound- the sounds of different vehicles once were rated when only the condition with whine was presented, and in another test with a direct comparison of the situations with and without whine (clutch engaged and disenga- ged). Differences in the ratings could be observed: it turned out that in the first case the overall whine was ra- ted, while in the latter case the attention of the listener was automatically focused to the gear whine, so that only that whine component was rated. But, even the rating of the overall whine impression was not stringent for all subjects. In a discussion it tur- ned out that a kind of global tonal impression arises, but that in addition usually one prominent tone was percei- ved caused by a single tonal component. But, different Fig. 1 Typical narrowband of an interior vehicle subjects detected different tones when listening to the sound (idle at 1200 rpm, red=right, green=left chan.) same stimuli. Furthermore, if this prominent tonal com- proofed chamber using headphone representation and ponent was eliminated, in a first instance the whine im- applying the individual test [3]. They have been based pression was reduced, but after listening to it for some on real interior vehicle recordings of several vehicles time, another tone seemed to „pop out“ of the back- where the existing tonal components have been modi- ground noise and could be heard as a single component. fied by filtering or synthetic sounds have been added. The perception of tonal components in these com- Psychoacoustical investigations have been conducted plex stimuli thus does not only depend on pure physical for the following parameter variations: parameters - it also depends on cognitive aspects, here · frequency variation of a single tonal component in the form of attention. The attention of a subject is fo- · combination of two tonal components cused on one of the components present in the stimuli. · combination of three tonal components This finding showed that most of the results of basic · a synthetic ideal sinusoidal tonal component perceptual investigations of tonal components can not · a real broad tonal component (bandwidth 80 Hz directly be applied to interior vehicle sound. In most ba- around 800 Hz). sic experiments the level of one tonal component was A group of experts and a group of non-experts parti- systematically varied, so that the attention of the sub- cipated in the tests and determined the detection and ac- jects was automatically focused to that component and ceptance thresholds and rated the strength of the whine. the contribution of other components was suppressed The findings can only be summarized here: by the listener. In a complex sound like the interior ve- · the expected frequency-dependency is reproduced: hicle noise the perceptual process is different, since the higher frequency components are more annoying attention of the subjects is not automatically focused. than lower frequency components (e.g., [2]) · the difference between detection and acceptance SUMMARY threshold depends on frequency and is about 6 dB at low and 3 dB at high frequencies The approach to adopt basic psychoacoustic data de- · experts render reproducible and stable results rived from experiments with single tonal components · non-experts have problems to rate whine in a re- by superposition of the effects can not be applied to the producible manner. The rating of a subset of sti- perception of multiple tonal components. In this com- muli rated two times depended on the other stimuli plex condition non-acoustic moderators influence the presented in the respective test (context effect) perception of humans. · broader components also causes tonal sensations comparable to sinusoidal components, but the LITERATURE spectral peak level is not the appropriate descriptor 1. Bodden, M., “Perceptual Sound Quality Evaluation“, in · the presence of multiple tonal components influ- Proc. Internoise 2000, Nice, France, 2000. ences the whine rating. It seems that contributions 2. Zwicker, E., and Fastl, H., Psychoacoustics -Facts and of several components are combined to produce a Models, Springer Verlag, 1990. whine rating. 3. Bodden, M., and Heinrichs, R., “Evaluation of interior ve- A very important and critical finding turned out in an hicle noise using an efficient psychoacoustic method“, in experiment where gear whine was investigated. There Proc. of the Euronoise 98, 1998. The influence of sound on perception thresholds and JNDs of whole-body

R. Weber, I. Baumann, M. Bellmann and V. Mellert

Department of - ACOUSTICS, Carl von Ossietzky University Oldenburg, D-26111 Oldenburg, Germany

For designing purposes regarding comfort, perception thresholds of vibrations are important for the expected effects of vibration. Just noticeable differences in level can indicate subjective efficiency of vibration changes. For a specific car seat, perception thresholds and just noticeable differences of whole-body vibration in the frequency range from 12 to 80 Hz are determined in the presence of two different noise levels. The seat is excited in z-direction. The thresholds are measured in using an adaptive 3 AFC 1up-2down method.

INTRODUCTION PERCEPTION THRESHOLDS

Whole-body vibrations usually have negative effects 12 subjects (6 male, 6 female / 22-32 years old) on subjective comfort perception which becomes participate in the adaptive tests (starting step size 8 dB increasingly important in transportation vehicles. Only / minimal step size 1 dB). The dashed curve in figure few basic research is published on the perception of 1 shows means of perception thresholds as a function whole-body vibrations in vehicles. Related norms as of frequency in the case of no additional acoustic e.g. ISO 2631 1/2 [3, 4], refer to whole-body signal. Results are given in terms of acceleration 2 vibrations in buildings. There subjects are seated on levels LVib in [dB] and acceleration a [m/s ] 2 rigid chairs that differently transfer vibrations (LVib. = 140 dB corresponds to a = 10 m/s ). compared to car seats. 100 0,1 This study investigates perception thresholds and just with pink noise without pink noise noticeable level differences (JNDs) of vertical whole- 95 0,0562 ²] s

body vibrations on a real car seat. Additional influence / 90

0,0316 m of (acoustical) noise on threshold is regarded. [ l.

85 0,0178 p

EXPERIMENTAL SETUP 80 0,01 acc level [dB]

75 0,0056 acc. am Using an adaptive 3-AFC 1up-2down-procedure 70 perception thresholds and JNDs of vibration levels 0,0032 1012,5 16 20 25 31,5 40 50 63 80 100 (70.7 % point of the psychometric function) are frequency [Hz] determined on a new car seat. Sinusoidal test frequencies vary from 12.5 Hz to 80 Hz in third octave FIGURE 1. Means and standard deviations of intervals. Signals of 1000 ms are seperated by pauses perception thresholds of vertically excited whole-body of 500 ms. Each threshold measurement is repeated vibrations on a real car seat (8 subjects) with (------) three times by each subject on three different days. and without (-----) masking (acoustic) pink noise of 68

Signal intervals are optically marked on a computer dB(A). monitor. The car seat is excited vertically by using a „sound and Starting with a level of 82 dB the perception threshold vibration reproduction system “, developed by the increases with a slope of about 5 dB/octave up to 50 ITAP GmbH in cooperation with the ACOUSTICS Hz whereas an unexpected decline is observed up to group [1, 5]. It is mounted on a decoupled basement in 80 Hz. a sound proof room with a acoustic background level To check whether sound radiated from the vibration of 38 dB(A). The device is optimised with respect to pad has triggered the subject’s responses and hence is diminish sound during vibration. responsible for the decline, tests are repeated using

pink noise - 50 Hz to 2 kHz / 68 dB(A) – as acoustic masker - presented via headphones (STAX). The perception thresholds of 8 subjects (6 males / 2 females) with masking noise ( line in figure 1) turn out to be on average higher by 1.4 dB. However, 3,0 they also exhibit this unexpected decline for ‘higher’ 2,5 vibration frequencies. Hence, this decline cannot 2,0 attributed to noise emitted by the vibrating pad. When comparing perception thresholds of Miwa 1,5 1969, McKay 1971, Parsons & Griffin 1988 (extracted 1,0 from [2]) and perception thresholds by Bellmann et al. 0,5

[1] with our own data essentially similar frequency level difference [dB] dependencies can be noted (see figure 2). Considerable 0,0 discrepancies exist with current norms [3, 4]. 12,5 16 20 25 31,5 40 50 63 80

Miwa, 1969 10frequency [Hz] 100 McKay, 1971 Parsons&Griffin, 1988 Parsons&Griffin, 1988 FIGURE 3. Means (thick solid) (std-devs included) 100 0,1 Bellmann et al., 2000 and medians (dashed) of just noticeable differences of ISO 2631 (z) vertically excited whole-body vibrations on a real car own data, without noise 95 own data, with noise seat (8 subjects) with a masking (acoustic) pink noise of 68 dB(A). Curve (thin solid) shows (rigid chair-) data from [1]. 90 0,0316

CONCLUSIONS

85 Perception thresholds of vertically excited sinusoidal whole-body vibrations (background noise level 38 dB) acc. level [dB] 80 0,01 on a real car seat start at 12.5 Hz with 82 dB and increase by about 5 dB/octave up to 50 Hz. At higher frequencies the perception threshold declines to 81 dB 75 at 80 Hz. Introducing an acoustic masking pink noise does not eliminate the decline of the perception threshold. It is on average about 1.4 dB higher than in 70 the ‘silent’ case. 1012,5 16 20 25 31,5 40 50 63 80 100 Just noticable differences of whole-body vibration frequency [Hz] level JND-Ls are 1.6 dB independent of frequency

from 12.5 Hz to 80 Hz. They are consistent with FIGURE 2. Comparison of literature data [1,2] on available literature results [1]. perception thresholds of whole-body vibrations and norm data [4] with own thresholds with and without REFERENCES masking noise.

1. Bellmann M.A., Mellert V., Reckhardt C. und Remmers JUST NOTICEABLE DIFFERENCES H., “Experimente zur Wahrnehmung von Vibrationen” In: Fortschritte der Akustik, DAGA 2000, 2000 10 subjects ( 6 male / 2 female) take part in 2. Griffin M. J., “Handbook of human vibration”, Academic Press, 1991 determining just noticeable differences of vibration 3. ISO 2631-1, “Evaluation of human exposure to whole- level. Starting level in the adaptive 3 AFC trials is 110 body vibration – Part 1: General requirements”, dB (= 0,316 m/s²), reference signal level is 100 dB International Organization for Standardisation, Geneva (= 0,1 m/s²), starting step size is 4 dB and minimal (1997) step size is 0.5 dB. 4. ISO 2631-2, “Evaluation of human exposure to whole- Resulting JND-Ls (figure 3) show frequency body vibration – Part 2: Continuous and shock-induced dependent means of about 1.6 dB with standard vibration in buildings (1-80 Hz)”, International deviations von 0.46 dB. Nearly no differences can be Organization for Standardisation, Geneva (1989) observed. The results fit well to former test results 5. Remmers H. & Bellmann M.A., “System zur realistischen Wiedergabe von Schall und Vibrationen”, obtained on a hard chair [1]. In: Fortschritte der Akustik, DAGA 2000, 2000

Psychoacoustic sensation magnitudes and sound quality ratings of upper middle class cars' idling noise

Ch. Patsouras a, H. Fastl a, D. Patsouras b, K. Pfaffelhuber b

a Institute for Human-Machine Communication, Technical University München, 80333 Munich, Germany, e-mail: [email protected] b FAIST Automotive GmbH & Co. KG, Krumbach, Germany

The outdoor idling noise of various upper middleclass cars - one gasoline powered car, four diesel powered cars of different brands, and three different adjustments of the motor of one diesel powered car - were assessed in psychoacoustic experiments. The relations between loudness, sharpness, roughness, fluctuation strength and the newly-defined sensation "dieselness" of those sounds will be discussed. It will be challenged which psychoacoustic sensations are instrumental for the preference of the sound quality of a specific car.

INTRODUCTION brand 1 ("dp b1") acted as anchor sound. The sound quality was assessed with a ranking method which is Standing next to a car, the idling noise of it is in most described in detail in [2]. Fourteen normalhearing cases a sufficient hint on the kind of motor. In contrast subjects with a median age of 27.5 years (4 female, 10 to a gasoline powered car, a diesel powered car shows male) participated in the experiments. a typical sound character which will be called in the following "dieselness". RESULTS To reveal the correlation between this characteristic sound described by dieselness and the basic psycho- Figure 1 shows the medians and the interquartile acoustic sensation magnitudes loudness, sharpness, ranges of the sensation magnitudes loudness (squares), roughness and fluctuation strength were investigated in sharpness (triangles), roughness (circles), fluctuation psychoacoustic experiments for eight different outdoor strength (rhombs) and dieselness (stars) for the idling noises. Among those eight cars were a gasoline gasoline powered car, the four different brands of powered car ("gp"), four diesel powered cars of diesel powered cars and the three different motor different brands ("dp b1" to "dp b4")as well as three adjustments of the diesel powered car. different adjustments of the motor of one particular diesel powered car ("dp a1" to "dp a3"). gp dp dp dp dp dp dp dp Furthermore, the hypothesis was posted that this b1 b2 b3 b4 a1 a2 a3 special sound character of diesel powered cars 250 (dieselness) is responsible for the judgement on the 200 sound quality of the car. Therefore, by means of a ranking experiment the sound quality of those eight 150 outdoor idling noises was assessed additionally. 100

50 EXPERIMENTS loudness sharpness roughness fluctuation strength

relative magnitude sensation / % 0 The outdoor idling noises of the above mentioned cars dieselness were recorded by a dummy head system of HEAD FIGURE 1. Results for the sensation magnitudes loudness, Acoustics positioned at a distance of 1 m lateral to the roughness, sharpness, fluctuation strength and dieselness. right front wheel at a height of 1.70 m. For the experiments, the sounds were presented in a sound- If the relative sensation magnitudes of the gasoline proof booth via a freefield equalized [1] STAX head- powered car are compared with those of the diesel phone calibrated to reproduce the original sound level. powered cars of different brands, it can be stated, that To evaluate the loudness, sharpness, roughness, the gasoline powered car produces about 65 % of the fluctuation strength and dieselness of the sounds, the loudness, 50 % of the sharpness, 60 % of the method of "magnitude estimation with anchor sound" roughness, 30 % of the fluctuation strength and only was used, and the sound of the diesel powered car of 10 % of the dieselness of that diesel powered car with correlation coefficients (table 2) between the rank in the respectively lowest estimated sensation magnitude. sound quality and the sensation magnitudes: in all Comparing the results of the four different brands, cases a strong correlation is given but especially the roughness is the sensation magnitude which is varying sensation dieselness (ρ = 1) seems to be an important most (about 50 percentage points) and fluctuation clue for the subjects in classifying the sound quality. strength less (about 15 percentage points). The dieselness is differing between the four different ran k 1 2 3 4 5 6 7 8 gp dp dp dp dp dp dp dp brands for maximum 32 percentage points. a2 b2 b1 a1 b4 b3 a3 With the investigated three adjustments of the motor, 250 fluctuation strength can be changed less (about 35 percentage points) and dieselness most (about 120 200 percentage points). In loudness, sharpness and roughness a difference of 85 to 100 percentage points, 150 in fluctuation strength of about only 35 percentage points can be obtained. 100 Table 1 shows the rank correlation coefficients (according to Spearman) between the sensation 50 loudness sharpness relative magnitude sensation / % dieselness and all other magnitudes. The correlation is roughness fluctuation strength 0 for all magnitudes very high but best between dieselness dieselness and roughness (ρ = 0,976). FIGURE 3. Results for the sensation magnitudes ordered TABLE 1. Rank correlation coefficients ρ between with respect to their sound quality. dieselness and the other sensation magnitudes loudness (N), sharpness (S), roughness (R) and fluctuation strength (F). TABLE 2. Rank correlation coefficients ρ between the rank in sound quality and the sensation magnitudes loudness (N), NS R F sharpness (S), roughness (R), fluctuation strength (F) and dieselness (D). 0,970 0,952 0,976 0,857 NS R F D Figure 2 shows the median and the interquartile ranges of the ranks given in sound quality for the eight cars 0,970 0,952 0,976 0,857 1,000 investigated. In judging the car with the best (gasoline powered car) and the worst (diesel powered car with CONCLUSION motor adjustment 3) sound quality subjects judged consistently. Whereas the diesel powered car with The outdoor idling noise of the gasoline powered car, motor adjustment 2 is rated better than all other diesel the diesel powered cars of different brands and powered cars, that one with motor adjustment 1 is especially the different adjustments of the motor of one classified behind brand 1 and 2 but still before brand 3 particular diesel powered car differ substantially in and 4. terms of the psychoacoustic sensation magnitudes 8 loudness, sharpness, roughness and fluctuation 7 strength. rank 6 The sensation characterizing the typical sound of a 5 diesel powered car - here called "dieselness" - is highly 4 correlated with those basic psychoacoustic sensation 3 2 magnitudes. Furthermore, the strength of the sensation 1 dieselness seems to be the cause how subjects rank the gp dp dp dp dp dp dp dp sound quality of the sound. b1 b2 b3 b4 a1 a2 a3 FIGURE 2. Results for the sound quality ratings. REFERENCES 1. Zwicker, E., Fastl, H., Psychoacoustics - Facts and DISCUSSION Models. 2nd updated ed., Springer Verlag, Berlin 1999. 2. Patsouras, Ch., Fastl, H., Patsouras, D., Pfaffelhuber, K., Figure 3 shows the results sorted with descending Subjective evaluation of loudness reduction and sound sound quality: an increase in all sensation magnitudes quality ratings obtained with simulations of acoustic seems to go in line with the deterioration in sound materials for noise control, in Proceedings of Euronoise 2001, edited by Demos Tsahalis, CD-Rom, 2001. quality. This can also be confirmed by the rank Effects of Modulation on the Quality of Diesel Engine Noise A. Hastingsa, P. Daviesa and H. Takatab aRay W. Herrick Laboratories, Purdue University, 1077 Ray W. Herrick Laboratories, West Lafayette, IN 47907-1077, United States of America bIsuzu Motors Ltd, 8 Tsuchidana, Fuijisawa-Shi, Kanagawa-Ken, 252 Japan

Diesel engine noise has strong tonal and time varying characteristics such as frequency and modulations and impulsiveness. Modulations in diesel engine sounds can be caused by variations in ignition timing, in the combustion pattern, and by piston slap. A model is constructed to synthesize diesel engine sound. The model is based on sequences of pulses that characterize combustion timing and amplitude in each of the engine’s cylinders. Measurements of cylinder during combustion are also incorporated. Deterministic and random variations in the timing and amplitude of the combustion processes were simulated at different levels to create a set of sounds with varying degrees of modulation. Subjective and objective (roughness and fluctuation strength) evaluations were used to examine the relationship between overall diesel engine sound quality and the degree of variability in timing and amplitude.

INTRODUCTION Impulse train

Diesel engines are simpler, have better fuel economy, The first step in synthesizing the sounds is to gener- and are more durable than gasoline engines. However, ate an impulse train for each cylinder that is based on people complain about diesel engine noise and often the engine rpm and the total number of cylinders (Nc). rate it as being more objectionable than gasoline engine The impulse train template for four stroke engines has a noise. Diesel engine noise may contain strong modula- periodicity dependent on the engine speed and the num- tions, tonal components and sound impulsive [1, 2, 3]. ber of cylinders, as shown in Equations 1 and 2, where Here the focus is on the effects of modulations. A model TSingleCylinder is the timing between consecutive impulses has been developed to synthesize diesel engine sounds in a cylinder and Teng is the time between impulses in the so that the relationship between modulations, caused by engine. timing and amplitude variations, and the sound quality of diesel engine noise can be examined.

120

T seconds ¡ (1) SingleCylinder rpm SYNTHESIS and T The main focus of the research described in this paper SingleCylinder Teng ¢ (2) is to gain a better understanding of how variation in com- Nc bustion timing and combustion pressure amplitude affect the perceived sound quality of diesel engines. Because The template has its amplitude and timing varied for it is difficult to separate the individual combustion events each cylinder. This variation has both a random and in a measured noise signature, these effects are being in- a fixed component. The fixed component causes each vestigated by synthesizing the sound of the combustion cylinder to be slightly different from the others. The ran- events starting with an impulse train for each cylinder in dom component causes individual combustion events to the engine. The main components of the synthesis are: be slightly different from the others, as might be the case • Generation of a no-variation (NV) uniform impulse if the combustion event is not truly repeatable and timing train for each cylinder based on engine revolutions cannot be tightly controlled. The fixed timing variation, starts by providing an off-

per minute (rpm). £

th ¥§¦ set so that the n cylinder fires n ¤ 1 Teng seconds after • Vary the timing and amplitude of each impulse train. the first. This time is then modified by a percent of Teng. • Convolve the impulse trains with combustion pres- The variation is again altered, this time randomly. This sure profiles. random variation has a uniform distribution with a maxi- mum level specified to be a particular percentage of Teng. • Filter the synthesized combustion to simu- The amplitude variation is done in a similar fashion, ex- late the relationship between combustion and acous- cept that there is no initial offset and the modifications are tic pressure. based on a percent of the template amplitude, not Teng. 1 3.3 0.85

0.9 3.2 0.8

0.8 3.1 0.75

0.7 3 0.7

0.6 2.9 0.65

0.5 2.8 0.6

0.4 2.7 0.55 0.3

Roughness Level, asper 2.6 0.5 Fluctuation Strength, vacil Pressure/max(Pressure), bar/bar 0.2 2.5 0.45 0.1 2.4 10% Variation 0.4 10% Variation 0 50% Variation 50% Variation 0 0.02 0.04 0.06 0.08 0.1 2.3 0.35 Time, seconds NV FA RA FT RT NV FA RA FT RT

FIGURE 1. The combustion impulse response. The pressure FIGURE 2. A comparison of modulation metrics with 10% and has been normalized to a peak amplitude of one. 50% variation. NV denotes No Variation. F and R denote the fixed and the random method of variation. A and T denote vari- ation applied to amplitude and timing. Combustion and system response

To convert an impulse train into a set of combus- tions tended to decrease roughness and increase fluctu- tion events, a combustion impulse response is used. A ation strength. To a lesser extent, amplitude variations single cylinder diesel engine’s combustion pressure was increased fluctuation strength, but did not have a strong recorded and averaged over 50 cycles. The result for the 1 impact on roughness. cylinder engine operating at 1200 rpm is shown in Figure The results of the metric calculations were confirmed 1. This averaged pressure profile was used as the basis for by the subjective analysis. When listening to the sounds, the combustion impulse response. Because this pressure it was found that sounds with the fixed variation applied profile changes with engine speed, the synthesis routine to either amplitude or timing had a stable modulated qual- compresses or expands this profile in time for higher or ity. When timing was varied, the modulations were per- lower, respectively, engine speeds. ceived to be much stronger than modulations resulting To convert the set of combustion events into actual from amplitude variations at the same level. Sounds with sound, a filter is needed to simulate the paths from the the random variations also had a modulated quality but cylinders to the listener. In the current study a simple sounded more erratic. Mirroring the objective analysis, lowpass filter is used and is the same for all cylinders. sounds with timing variations sounded less rough than the Future refinements of the simulation will include more original sound with no variation. realistic models of these paths.

ACKNOWLEDGMENTS SOUND QUALITY ANALYSIS The Authors would like to thank Isuzu Motors for their In order to determine how timing and amplitude varia- sponsoring of this research and Lijun Song and Dr. John tion effect sound quality, sounds were generated with dif- Abraham of the Zucrow Laboratories for their assistance ferent levels of fixed and random variation applied to the in acquiring initial combustion profiles. amplitudes and timings of the impulse trains. These vari- ations ranged from zero to fifty percent of the amplitude and the Teng values. The rpm of a six cylinder engine was REFERENCES set at 700, making the nominal no-variation timing be- tween engine combustion events 0.0286 seconds. After 1. R. Ingham, N. Otto and T. McCollum, The Society of Auto- all sounds were adjusted to the same loudness level, met- motive Engineers. 1999-01-1819, 1295-1299 (1999). rics were calculated and the sounds were also evaluated 2. H. Takata, T. Nishi and P. Davies, Proceedings of Inter- subjectively. Noise 99, Fort Lauderdale, Florida, 1201-1206 (1999) It was found that the roughness and fluctuation 3. M. Russell, S. Worley and C. Young, The Society of Auto- strength metrics were sensitive to the type as well as motive Engineers. 870958, 79-95 (1987). the level of the variation. Example results for 10% and 50% variations are shown in Figure 2. Timing varia-

Effect of Factors other than Sound to the Perception of Sound Quality

T.Hashimotoa and S. Hatanoa

aDepartment of , Seikei University, 3-3-1 Kichijoji Kitamachi, Musashino, Tokyo 180-8633, Japan

For the evaluation of sound quality of car interior noise, factors other than sound, e.g., image of a car, seat vibrations, scenery from the car etc., affect the results. In order to examine these effects, evaluations of car interior noise with the simultaneous exposure of visual image were conducted. Besides this, the evaluation only with a car image without noise exposure was also conducted. As a result, the evaluation obtained only by an image was usually the best and that obtained by a noise exposure was the worst, and the latter became better with simultaneous exposure of noise and image and this was in between the two extremes. If we swap the image of the cars to the real one, this also affected the sound quality evaluation. For example, the sound quality of an expensive car reduced the ratings due to the simultaneous exposure of image of a cheap car and vice a versa. Together with the effect of image, seat vibration of a car affected the sound quality evaluation on pleasantness, powerfulness and booming sensation of the sound. If the seat vibration was simultaneously exposed to the subjects, they responses to the sound were more intense compared with the case with no vibration.

INTRODUCTION

The evaluation of sound quality of car interior noise is The height of the moving image also affected the affected by various parameters such as visual scenery, subjective evaluation of car interior noise although the image of a car, seat/floor vibrations etc[1,2]. In order result was not shown here because of the shortage of to see these effects quantitatively, two types of the space. subjective evaluation test were conducted using SD Table 1 List of adjectives for SD experiment method. The one is to see the effect of image a car and athin-thick the other is to see the effect of visual scenery and b unsatisfactory -powerful seat/floor vibrations to the perception of sound quality. c booming - ringing d cheap - expensive e unpleasant - pleasant SD EXRERIMENT TO SEE THE f clamorous - quiet EFFECT OF IMAGE OF A CAR gdull-sharp hmuddy-clear itight-loose The image of a car affects significantly the result of j rough - smooth subjective evaluation of sound quality of car interior khard-soft noise. In order to see this effect, SD experiments were l shrill - calm m heavy - light conducted.

The effect of an image of a car was tested by SD 7 by image method. Subjects were 20 males and 20 females aged by sound 6 by image and sound between 18 and 25 years. As was shown in Fig.1, the 5 evaluation only by moving image was the best and that 4

obtained only by noise exposure was the worst and 3 evaluation that obtained by simultaneous exposure of moving 2 image and noise was in between the two extreme. 1 abcde f gh i j k lm Another result for the same subjects revealed that the effect of simultaneous exposure of visual image was sometime equivalent to the reduction of 10dB SPL in FIGURE 1. Effect of moving scenery from the car the case where only the interior noise was exposed. inside to the perception of sound quality

The effect of simultaneous exposure of noise and seat/ 7 floor/ steering wheel vibrations to the perception of 6 unpleasantness and the effect of simultaneous 5 exposure of noise and scenery were examined by 4 comparing the result obtained only by noise exposure. 3 As was seen from the Fig. 4, the effect of vibrations

evaluation strengthened the unpleasantness and the effect of by sound 2 by sound and image scenery weakened the unpleasantness. by sound reduced 10dB 1 a b c d e f g h i j k l m0 7

FIGURE 2. Effect of moving image from the car 6 inside was equivalent to 10dB reduction in SPL. 5 EFFECTS OF SEAT/FLOOR VIBRATIONS AND SCENERY 4

3 In order to see the effect of seat/floor/steering wheel Noise vibrations together with the moving scenery from the pleasant-unpleasant 2 Noise+Vibration car inside, laboratory test using the test facility shown Niose+Scenery in Fig.3 was conducted. The vibrations were initially 1 recorded in a real car at the positions of the seat, floor and steering wheel under the real running condition 0 and these signals were reproduced in a sound proof 0 100 200 300 400 500 time (sec) room together with the interior noise and the moving scenery. The subjects joined were 19 males and 1 female aged between 22 to 55 years with normal FIGURE 4. Effect of vibration and scenery to the hearing. perception of unpleasantness of car interior noise Instantaneous evaluations on three factors namely, pleasantness, powerfulness, and booming sensation, were collected using computer keyboard as the input device. The program for collecting subjective CONCLUSIONS responses was made using visual Basic . 1. Effect of visual image reduces the negative impression of sound quality and the amount is seat sometime equivalent to 10dB reduction in SPL.

Projector 2. Seat/floor/steering wheel vibrations strengthen the headphone unpleasantness while the scenery reduces the Screen unpleasantness. transducer Steering wheel

Speaker Woofer REFERENCES

1. T.Hashimoto., Proc.JSAE, No.30-00,13-16 (2000), (in Japanese with English summary).

2. T.Hashimoto., Proc. JSAE Noise and Vibration Exciter Spring Forum, 1-4(2001), (in Japanese with English summary) . FIGURE 3. Test facility

Inter-Modal Effects of Non-Simultaneous Stimulus Presentation M.E. Altinsoya, J. Blauerta ,C. Treierb aInstitut für Kommunikationsakustik, Ruhr-Universität Bochum, D-44780 Bochum, Germany bInstitut für Arbeitswissenschaft, Ruhr Universität Bochum, D-44780 Bochum, Germany

In environments where products are made use of and the product noise interferes with these activities (e.g. in the car, in an air plane, or when operating a machine or an appliance) non-acoustics factors like vibration, heat, visual effects may heavily influence the judgements on product-sound quality. Auditory virtual-reality generators are potent tools for psycho-acoustic research in complex, interactive auditory scenarios. They can be extended to serve as a tool for multi-modal psychophysics by including non-auditory modalities. Such multi-modal virtual-reality generators become more and more important as tools for product-sound-quality evaluation. However, the integration of further modalities may create problems. To gain a better under- standing of the integration of auditory, visual and haptic information, it is necessary to specify which criteria have to be met with respect to temporal factors, particularly synchrony. The objective of this paper is to provide an overview on this topic. 1. INTRODUCTION perceptual realism provided by VE Systems. Miner and Caudell’s computational analysis reveals that acoustic In an earlier paper, we have defined product-sound processing delays of at least 66 ms must be expected quality as “a descriptor of the adequacy of the sound with today’s technology for producing a rather attached to a product. It results from judgements upon simplistic auditory field. As the complexity of auditory the totality of auditory characteristics of the said sound environments increases, the computation time and - the judgements being performed with reference to the resources will also do so. Audio delays of this set of those desired features of the product which are magnitude may have a negative impact on several apparent to the users in their actual cognitive, actional aspects of VE simulations [6]. Barfield et al. compared and emotional situation” [1]. In user interaction with the human’s ability to detect and discriminate visual, complex products various information reaches the user auditory, tactile and kinesthetic information with the from different modalities. Consequently; the cross- current technical specifications of virtual-environment modal information is of substantial influence when the equipment, but for each modalities seperately. Their evaluating of the product-sound quality. Auditory vir- study does not include the human ability of multi-modal tual-reality generators have proved to be potent tools information detection and discrimination [7]. A multi- for psychoacoustic research in complex, interactive modal synchronisation threshold can be defined as the auditory scenarios and there are strong efforts recently maximum tolerable temporal separation of the onset of to use them as car and aircraft simulators for product- two stimuli, one of which is presented to one and sound-quality research [2,3,4]. One of the major ob- the other to another sense, such that the accompanying jectives of virtual-environment (VE) designers and sensory objects are perceived as being synchronous. In researchers is to obtain more realistic and compelling order to measure this threshold, observers may be asked virtual environments. This objective, though, requires a to report which of the two stimuli comes first (forced better understanding of the integration of the main sen- choice). In multi-modal-interaction research there are sory modalities, namely, auditory, visual and tactile. An several studies regarding the detection of understanding of perceptual aspects of temporal factors synchronisation thresholds. Most investigations which on multi-modal presentations is very important to deal with the effect of temporal factors in multi-modal determine how to integrate multi-modal information. presentation are related to audio-visual synchrony, This paper presents some overview of temporal aspects while only very few investigations address auditory- of multi-modal integration, based on literature data. tactile synchrony. The obtained results vary, depending on the kind of stimuli and the psychometric methods 2. TEMPORAL ASPECTS OF MULTI- employed. Hirsh and Sherrick measured the MODAL INTEGRATION synchronisation thresholds regarding visual, auditory and tactile modalities [8]. They presented 666 Hz, 10- Multi-sensory integration has been defined by Kohl- ms sine pulses as acoustic stimuli via headphone, rausch and van de Par as the synthesis of information similar pulses as tactile stimuli to the tip of the index from two or more sensory modalities such that infor- finger via a shaker, as well as 5-ms flashes of light on a mation emerges which could not have been obtained screen as visual stimuli. The subjects were asked to from each of the sensory modalities seperately [5]. report which stimulus came first. Their results show Effects of non-synchronisation on perception (i.e.. that the visual system is relatively sluggish while the delays between modalities) are important factors for the tactual system is less so, and the seems to be fastest (see Table 1). Dixon and Spitz investigated and constant stimuli without feedback ) were used and synchronisation-threshold differences between two obtained two different results. The authors report that different stimuli, namely, and hammer-strikes the synchrony curves obtained are not centred around [9]. They found that the synchronisation threshold of an AV delay of 0 ms. The reason for this may be speech is higher than that for the impact stimuli. Miner different time spans needed for the internal processing and Caudell conducted a cross-modal psycho-acoustic of auditory and visual stimuli. Obviously the human experiment to measure perceptual perceptual system is adapted to tolerate large audio delays which, consequently, may result in lower Table 1. Synchronisation-threshold values sensitivities to audio delays as compared to video delays. Further, the experimental procedure chosen has Hirsh & Sherrick a strong influence on the results. audio visual audio delay visual delay 20 ms 20 ms audio-tactile audio delay tactile delay 3. ONGOING RESEARCH 25 ms 12 ms visual-tactile visual delay tactile delay Many virtual auditory-tactile environments, e.g. car 20 ms 30 ms simulators, require a whole-body tactile stimulation. Dixon & Spitz For this case, literature data are not available. audio-visual audio delay visual delay hammer-impact 188 ms 75 ms Therefore, further investigations on the auditory-tactile speech 258 ms 131 ms synchrony are currently carried out by us to measure the Miner & Caudell synchronisation threshold of auditory-tactile pres- audio-visual audio delay entations using a whole-body vibrator and a virtual hammer 177,17 ms drum Stick 176,70 ms auditory environment. Both realistic and artificial glasses 191,15 ms stimuli are used. In the first case an auditory-tactile Newton’s cradle 172,70 ms recording was made of a car passing a bump. In the speech 203,32 ms overall mean 184,21 ms second case a broad-band noise for the auditory Lewkowicz stimulus and a sine for the tactile stimulus were audio-visual audio delay visual delay employed. During the experiments, tactile and auditory adult 112 ms 65 ms stimuli are presented with various inter-stimuli delays. infant 450 ms 350 ms Kohlrausch & van de Par The results of this investigation will be ready for oral audio-visual audio delay visual delay presentation at the congress. AFC method 85 ms 29 ms constant. meth. 175 ms 75 ms 4. REFERENCES

audio-visual-synchronisation thresholds for audio- [1] Blauert, J., Jekosch, U., 1997. ACUSTICA/acta acustica, 83, signal delay [6]. Three different single impact events (a 747-753 dead-blow hammer striking a lead block, two wine [2] Genuit, K., 2000,.“The future of sound quality of the interior noise of vehicles”, in: Proc. Internoise 2000, 1693-1698, F-Nice, glasses colliding, and two drumsticks striking), one [3] Quehl, J., Schick, A., Mellert, V., Schulte-Fortkamp, B., Rem- repeated-impact event (two suspended silver balls mers, H. 2000. “Dimensions of Combined Acoustic and Vibration colliding twelve times, Newton’s cradle), and speech Perception in Aircrafts Derived by Factor Analyis of Semantic segment were selected to determine the threshold Differential Data”, in Proc. Internoise 2000, 465-469, F-Nice [4] Hillebrand, P., Schaaf, K., 2000. “Anwendungen eines vibroa- variation for different characteristic stimuli. The kustischen Simulators in der Automobilindustrie”, in Fortschr. experimental results suggest a limited audio-processing Akustik, DAGA 2001, Dtsch. Ges. Akustik, D-Oldenburg budget available in terms of audio-visual synchro- [5] Kohlrausch A., van de Par S., 1999. “Auditory-visual interac- nisation requirements. Lewkowicz performed an ex- tion: From fundamental research in cognitive to periment to investigate the AV-asynchrony-threshold (possible) applications”, in: Human Vision and Electronic Imaging IV, Proc. Soc. Photo-Optical Instrumentation Engrs. 3644 , 34-44 differences between adults and infants. Participants [6] Miner N., Caudell T., 1998. Presence 7, 396-409 (1998) were familiarised with a bouncing disk and a sound that [7] Barfield, W., Hendrix, C., Bjorneseth, O., Kaczmarek, K.A., occurred each time the disk bounced. Then they were Lotens, W., 1995. Presence 4, 329-356 given a series of asynchrony test trials where the sound [8] Hirsh I.J., and Sherrrick C.E, 1961. J. Exp. Psychol 62, 423- 432 occurred either before or after the disk bounced [10]. Dixon N.F., Spitz L., Perception 9, 719-721 Sensitivity to auditory-visual asynchrony was measured [10] Lewkowicz, DJ, 1996. J. Exp. Psych. 22, 1094-1106 by van de Par and Kohlrausch [11] using brief tonal [11] Van de Par S., Kohlrausch A., 1999. IPO Ann. Progr. Rep.34, signals with a frequency of 500 Hz, accompanied by a 94-102 moving disk on a monitor which was visible for 2000 ms. In this study two different psychophysical measuring methods (alternative forced choice, AFC, Psychoacoustic Correlates of Time and Spectral Characteristics of Railway Noise

A. Preis, R. Golebiewski

Institute of Acoustics, Adam Mickiewicz University, 61-614Poznan, Poland

The authors report how the annoyance assessment of the railway noise depends on the and the distance of the passing train. Annoyance judgments of a moving train cannot be merely explained on the basis of the spectral content and sound level of noise source. Loudness spectra and time patterns of the noise of a train moving at the same velocity but recorded at different distances deliver important cues that help to account for the differences in annoyance judgements. The autocorrelation function calculated on the time patterns of all stimuli allows to identify periodic components occurring in the stimulus. The paper aims to determine to what extent spectral and time characteristics of railway noise contribute to its annoyance.

INTRODUCTION source were used as test stimuli. There were noises generated by Inter City (IC), passenger (PT) and Noise produced by source located farther from a goods trains (GT), each of 25s duration. The velocity listener is assessed as less annoying than noise of each train is presented in Table 1. produced by source of the same type that is located closer. This is explained by pointing to the fact that Subjects increase in distance of the source of noise causes decrease in loudness. However, the ground effect and Subjects were normal-hearing students of the air phenomenon change not only the sound A. Mickiewicz University (15 female and 16 male). level of the stimulus but also its spectral content. The aim of the present paper is to answer the Procedure questions: (1) what is the annoyance of noise produced by similar sources located at different distances but Subjects were given the following instruction: “You reaching listener's with the same loudness? (2) to will listen to a pair of railway noises, please mark what extent spectral and time characteristics of railway which of the two noises presented in a pair you would noise contribute to its annoyance? prefer to switch off, if given the possibility”. Each In the present , noise of a train moving at subject judged each pair of noises only once. The different , recorded at two distances was the results are presented as the percentage of occurrences object of annoyance judgments. The original 12 of a particular noise chosen as non-preferred railway noises recorded at two distances were component in a pair to all occurrences of this noise. artificially modified to make them equal in loudness. The higher this non-preference percentage, the greater The loudness equalization was done in two steps. At was the number of subjects who wanted to switch this first, the sound recorded at the farther distance was particular noise off. In all, each of the 31 subjects amplified in a linear way until its LAE was the same as made 24 preference judgements, one for each of 24 the LAE of the noise recorded at the closer distance pairs. from the source. Then, based on the spectra of these noises the loudness, N, according to the Zwicker RESULTS method [3] was (ISO532B) calculated. This procedure guaranteed that the noises presented in pairs were The results of these comparisons are presented in the always equal in LAE but not always in their loudness. Table 1. The pairs of noises equal in loudness are In the psychoacoustic experiment, subjects were asked marked in the Table 1 with an asterisk. We assume which of the two noises presented in a pair they would that results above 75% indicate that subjects’ choices prefer to switch off given such a possibility. were not casual. It can be seen that subjects’ choices of Stimuli and apparatus the more annoying component are in agreement with the lower frequency of the periodic component found Twelve original railway noises, recorded at two in the autocorrelation function calculated for all noises. distances (S1=25m and S2=450m) from the moving Periodic component is defined by the time delay at the first maximum peak of the normalized autocorrelation r function [1] (it corresponds to the phenomenon of the 4,50 virtual pitch [2]). It seems, that periodic components of 4,00 3,50 PT-ORG noise generated by IC train were too high to favor of 3,00 PT-MOD the choice of the more annoying component. 2,50 2,00 FIGURE 1. Averaged loudness spectra for original and 1,50 1,00 modified PT train noises. 0,50 0,00 Specific loudness[sone/ba Specific

5 5 ,5 5 5 ,5 5 5 0, 3, 6 9, 2, 1 15 18, 21, Barks Table 1. Percentage of occurrences of noise chosen as non-preferred component in a pair (column s 2 and 3), frequency of periodic component (columns 4 and 5), velocity of the train (column 6).

Train S1-org[%] S2-mod[%] S1-org S2-mod Velocity, V Period. Period. comp. [km/h] Comp.[Hz] [Hz] *IC1 42 58 1818 800 126 *IC2 48 52 no peak 1000 140 IC3 58 42 2000 909 134 IC4 58 42 2000 952 138 PT1 84 16 357 1250 97 *PT2 87 13 312 830 95 PT3 90 10 263 833 95 *PT4 87 13 357 769 95 GT1 23 77 312 140 76 GT2 77 23 136 833 68 *GT3 81 19 125 144 60 *GT4 48 52 126 no peak 76

90,0 80,0 . If it is true, then annoyance judgements 70,0 of noise produced by the moving source depend on 60,0 source’s velocity. 50,0 40,0 30,0 PT-ORG 20,0 10,0 PT-MOD REFERENCES

Sound exposure level [d 0,0

0 0 0 0 0 0 0 1. Y. Ando, H. Sakai and S. Sato., Journal of Sound and 50 10 20 40 80 60 1 315 630 Vibration 232, 101-127 (2000). Frequency [Hz] 2. E. Terhardt., Hearing Research 1, 155-182 (1979). FIGURE 2. Averaged spectra of L AE, for original and modified PT train noises . 3. E. Zwicker and H. Fastl., Psychoacoustics-Facts and Low frequency components occurred in original noise Models, Heidelberg; Springer-Verlag, 1990, pp. 289-291. generated by PT trains. Noises with these low frequency components were chosen as distinctively more annoying. These low frequency components can be seen in loudness spectra in Figure1. However, they are not visible, when we present the same stimuli as the averaged spectra expressed in dBA (see Figure 2).

The differences in annoyance judgments of IC train noise compared to the PT and GT train noise may be explained by pointing to the higher velocity of IC trains. This difference might be the result of the Onomatopoeic Features of Sounds Emitted from Laser Printers and Copy Machines and Their Contribution to Product Image M. Takada1, K. Tanaka1, S. Iwamiya1, K. Kawahara1, A. Takanashi2 and A. Mori2 1 Dept. of Acoustic Design, Kyushu Institute of Design, 4-9-1 Shiobaru, Minami-ku, Fukuoka, 815-8540, JAPAN 2 CANON INC., 3-30-2 Shimomaruko, Ohta-ku, Tokyo, 146-8501, JAPAN

In this study, to clarify the acoustical properties of the manual operation sounds of laser printers and copy machines, psychoacoustical experiments using onomatopoeic representation were examined. The similarities of phonetic parameters in the onomatopoeic representations among sound stimuli were applied to the hierarchical cluster analysis. As a result, the sound stimuli were categorized into 5 clusters, according to the acoustical properties in the and the time domain. Furthermore, the product images associated with its sounds were measured by a rating experiment. Several relationships between the product image and phonetic features were clarified: for example, the prolonged sound of /i/ and voiced consonants are associated with unpleasantness.

INTRODUCTION the free description experiment using onomatopoeic representations for the manual operation sounds. The Recently, in offices, there are many kinds of sounds other was the measurement of the impression of the same from office equipment, such as computers, printers and sound stimuli using a 7-step adjective scale of “pleasant – copy machines. Printers and copy machines emit various unpleasant” and “strong – breakable”. Fourteen native types of manual operation sounds when we open and speakers of Japanese (6 male and 8 female subjects) close top covers, draw and load paper trays, and lock and participated in these experiments. release hooks. These sounds contribute to the quality of products and the reliability of operation. Users often feel DISCUSSION anxiety from some types of manual operation sounds. When users complain about manual operation sounds Analysis of Onomatopoeic Representations to distributors, onomatopoeic representations are To clarify the onomatopoeic features to represent the frequently used. The onomatopoeias are a natural way to acoustical properties of the manual operation sounds, the express auditory sensations. They may reflect the acoustic onomatopoeic representations obtained were coded using features which affects the auditory impression and 24 phonetic parameters such as 7 places of articulation, 6 imagery. The onomatopoeic representations can be used manners of articulation [1], 5 vowels in Japanese, voiced to measure product quality without actually doing the and voiceless consonants, syllabic nasal, choked sound, formal psychoacoustical experiment. palatalized sound and prolonged sound. Furthermore, The psychoacoustical experiments were conducted to cluster analysis was applied to the similarities among the investigate the possibility of using onomatopoeic repre- sound stimuli expressed by the frequencies of the 24 sentations to estimate the quality of laser printers and phonetic parameters. The result of cluster analysis is copy machines. Firstly, the free description evaluation shown in Figure 1. The type of stimuli, examples of experiment using onomatopoeic representations was onomatopoeic representations and the average number of examined for manual operation sounds. Furthermore, phonemes are shown in each cluster. All stimuli were subjective evaluation experiments using adjective scales categorized into 5 clusters. on the product image were examined. Then, the The sound stimuli in cluster 1 and cluster 2 consist of relationship between the product image and the onomato- rubbing sounds from the operated part and impulsive poeic features was discussed. striking sounds. The rubbing sounds in cluster 2 have especially high energy in the high frequency region. In EXPERIMENTAL METHOD onomatopoeic representations for these “sharp” sounds, the Japanese vowel of /i/ [i] and its prolonged sounds are Forty four manual operation sounds, such as opening frequently used such as /kiiiii/ [ki] (/keeey/ in English and closing front covers, drawing and loading paper trays, and locking and releasing hooks of 3 laser printers and 2 expression). The Japanese vowel of /i/ [i] with the highest copy machines were used as the sound stimuli. Most of second frequency (about 2.8kHz) in the vowels is them have impulsive striking sounds emitted from the used in order to represent the sharp impression [2]. The operated parts such as front covers, paper trays and hooks. rubbing sounds in cluster 1 have energy not only in the They were recorded in an office-like room using a high frequency region but also in the lower frequency dummy head recording system. These stimuli were region. In the onomatopoeic representations, the other presented to subjects from a computer via headphones. vowels, except /i/ and their prolonged sounds, are often Two kinds of experiments were conducted. One was used such as /shuuuu/ []. 1: locks, trays, covers sound. Furthermore, the rank correlation coefficients be- [s bai],[ bata] tween the number of each phoneme and average pleasant- phonemes: 5.4 ness scores were calculated. In the group of unpleasant sounds, there are all stimuli 2: covers having the “sharp” rubbing sound before the striking [ki ba],[kji ta] sound from the operated part, as shown in Figure 2. It is phonemes: 6.0 supposed that these stimuli were evaluated to be un- 3: covers pleasant due to the “sharp” rubbing sounds. These sharp [ba], [k], [da] and unpleasant sounds are represented by consonants with phonemes: 3.5 the vowel of /i/ and its prolonged sounds such as /kiiiii/ [ki] and /giiiii/ [i] (/geee/ in English expression). The 4: trays correlation coefficients between the average pleasantness [do kaaado tontontonto ] phonemes: 8.7 scores and the number of phonemes /i/ [i], /hi/ [i] and /kyu/ [kj] are statistically significant at a level of 0.01 5: covers, sliders, locks (r=0.439, r=0.417 and r=0.445 ). [aaa ta],[ktatata ta] The beginnings of the onomatopoeic representations for phonemes: 5.0 The number of cluster:type of stimulus the impulsive striking sounds emitted from the operated Examples of onomatopoeic representations parts are expressed by the stops, such as /ban/ [ba], /pan/ Average number of phonemes [pa] and /ga tan/ [ta]. Especially, for the un- FIGURE 1. Dendorogram of the manual operation sounds pleasant sounds, the stops of voiced consonants such as /g/

The sound stimuli in cluster 3 have only striking sounds. and /b/ (for example, /ga/ [], /ba/ [ba] and /be/ [be]) are Therefore, the onomatopoeic representations for them are used at the beginnings of the striking sounds more shorter than those in the other clusters. The average frequently than those for the pleasant sounds. The spectra number of phonemes in the onomatopoeic representations of the voiced consonants have higher energy in the high is 3.5. Examples of the onomatopoeic representations for frequency region than that of the voiceless consonants [3]. these stimuli are simple, such as /ban/ [ba], /kan/ [k] Therefore, to represent the “harsh” striking sounds with high energy in the high frequency region, the voiced con- and /dan/ [da]. sonants should be used. The correlation coefficients Two sound stimuli categorized in cluster 4 are the long- between the average scores on pleasantness and the est in duration. They have continuous sounds emitted from the operated parts after striking. Therefore, the onomat- number of phonemes such as /ga/ [], /be/ [be] and /gwa/ opoeic representations for them are longer than those in [a] are statistically significant at a level of 0.01 other clusters. The average number of phonemes in the (r=0.406, r=0.453 and r=0.393). The correlation onomatopoeic representations is 8.7. As an example of the coefficients between the average scores in the image of onomatopoeic representations for the continuous sounds, strength and the number of phonemes such as /ga/ and repeated syllables are used such as /ton ton ton/ /be/ also are statistically significant. 20 [tontonto]. 10 The sound stimuli in cluster 5 have many short sounds 5 successively emitted from the operated parts before 2 striking. There are short pauses between many short 1 sounds. To represent the short sounds with the short 0.5 pauses, the stop and the flapped articulated in the alveolar Sharp rubbing sound 0.2 are often used such as /ka ta ta ta/ [ktatata]. 0. 1 Harsh striking sound 0.05 From these results, the onomatopoeia is a valid way to capture the acoustical properties of the manual operation 200ms 0.02 0.01 sounds in the time and frequency domains. Time [s] [kHz] FIGURE 2. Wavelet analysis of the most unpleasant sound The Relationships between Product Image with a “sharp” rubbing sound and a “harsh” striking sound and Onomatopoeic Features REFERENCES To clarify the relationships between the product image, 1. H.Jyouo, “Phonetics in Japanese”, Bandai Music Entertainment, Tokyo, such as pleasantness and strength associated with sound 1998 (in Japanese) stimuli, and the onomatopoeic features, the number of 2. S.Iwamiya and M.Nakagawa, “Classification of audio signals using whole phonemes in the onomatopoeic representations of onomatopoeia”, , Vol.2, 23-30 (2000) (in Japanese) 3. K.Tanaka, K.Matsubara and T.Sato, “Onomatopoeia expression for all subjects for each sound were counted in each period of strange noise of machines”, J. Acous. Soc. Japan, Vol.53 (6), 477-482 the striking sound, the previous sound and the following (1997) (in Japanese) Perception of directional pitch change observed in monkeys and humans

H. Riquimaroux, T. Takahashi and K. Sumida

Department of Knowledge Engineering and Computer Sciences, Doshisha University, Kyotanabe, Kyoto 610-0321, Japan

A pair of tone bursts and/or harmonically structured complex tone bursts were sequentially presented on directional pitch change discrimination tasks for Japanese monkeys and human subjects. Results show that monkeys and humans may have similar sequential pitch. However, both may have common difficulty in judging the pitch direction by using the cue when the temporal sequence is made of a simple tone burst and a complex tone burst where the frequency of the simple tone burst is in between the fundamental and the second harmonic frequencies of the complex tone burst. Results suggest that the music training could make a difference in judging what cue should be used in human subjects.

INTRODUCTION was presented, they could drink water from the spout. Pushing the button during negative stimulus (S-) was We have neurophysiologically and behaviorally punished with a time out. Sound stimuli were investigated pitch extraction system in the Japanese presented from a loud speaker fixed in front of the monkey [1,2,3]. Our behavioral studies have revealed animal. Stimuli were sequentially presented simple that Japanese monkeys show similar pitch perception tone bursts and harmonically structured complex tone that humans do including the bursts (Figure 2). Figure 2a indicated envelope pattern perception[2]. In this research perception of directional of two sequential tones. For Subject M1, S+ was pitch change in monkeys and humans are compared. defined by an increment of the fundamental frequency between the two tones while S- was defined by the METHODS opposite direction. Schematic of a set of

Two Japanese monkeys (Macaca fuscata), Subjects M1 (4 years old) and M2 (5 years old) were used. They sat in a monkey chair equipped with a steel spout, an infrared sensor and a push button (Figure 1). If the subject pushed the button while positive stimulus (S+)

FIGURE 2. Envelope pattern of sequential tones. (a) For monkeys the duration of both tone bursts was always 200 ms. For humans the duration of a tone burst was varied, but the duration of tones 1 and 2 was always the same. (b) This patterns was used only for human subjects. The interval between two tones was varied. FIGURE 1. Apparatus. fundamental frequency lower than 300 Hz persistently. Positive stimuli (S+) Negative stimuli (S-) These reversal responses were often found when the frequency ratio between the first and the second tone was small. Human subjects also sometimes responded in the same way for the same stimulus as used to monkeys. So, for human subjects additional conditions were conducted. Musically experienced subjects often showed no error performance, judging Frequency with the change in the fundamental frequency. Others often showed improvement as tone duration lengthened or interval between two tones lengthened Time (Figure 4).

FIGURE 3. Schematic spectrograms of a set of tones

  

for Subject M1. 

6 S1

    

     tones for Subject M1 were indicated in Figure 3. For 5  S2

     Subject M2, S+ and S- were switched. T1 was a pair of 4  S3

    

tone bursts. T2 and T3 were pairs of complex tones. 3  S4

   

T4 and T5 were combination of a tone burst and a 2 

responses    



complex tone. The frequency of the simple tone burst 1

    was in between the fundamental frequency and the  0 second harmonic of the complex tone burst for T4. The fundamental frequency of the complex tone burst 0 200 400 800 was higher than the frequency of the simple tone burst interval (ms) for T5. T6 and T7 were a pair of tone bursts where the highest were the same frequency. The fundamental frequency was selected randomly from FIGURE 4. Number of correct responses at each 150 – 1000 Hz. The frequencies of the highest interval between tones for human subjects. harmonics were lower than 9000 Hz. The frequency ratio between the first and the second tones in a sequence was 1.25 - 1.89. level was ACKNOWLEDGMENTS randomly varied between 30 and 50 dB SPL. The stimulus set was replaced with an unfamiliar set each This research was supported by Special Coordination session. In order to compare data obtained from Funds from the Science and Technology Agency and a humans, seven human subjects were used in the same grant to RCAST at Doshisha University from the paradigm except of the water reward. For human Ministry of Education and Science of Japan. subject, after habituation training to the experimental condition with the same stimuli used for monkeys, T4 and T5 patterns with different temporal structure were presented. Duration of tones was varied from 200 to REFERENCES 1000 ms (Figure 2a), or interval between two tones was varied from 0 to 800 ms (Figure 2b). 1. H. Riquimaroux, S. Toriyama and K. Manabe, “Sequentially perceived pitch in Japanese monkey,” in Proc. of the International Symposium on Recent Developments in Auditory Mechamics, edited by H. Wada et al., World Scientific, 1998, pp. 450-456. RESULTS AND DISCUSSION 2. K. Manabe, S. Toriyama, and H. Riquimaroux, Proc. The monkeys were successfully trained to Autumn Meet. Acoust. Soc. Jpn. 457-458 (1997) discriminate the direction of fundamental frequency and/or pitch change, rising or falling in the same 3. N. Kitagawa, S. Toriyama, and H. Riquimaroux, Proc. manner as our precedent experiment with the correct Spring Meet. Acoust. Soc. Jpn. .381-382 (2000) ratio above 90 % [4] except for the tone pattern T4 [5]. For T4 pattern, correct ratio of Subject M1 was 70 % 4. N. Kitagawa and H. Riquimaroux, Trans. Tech. Comm. and that of Subject M2 was 80 %, and they tended to Psychol. Physiol. Acoust. , Vol. H-2000-31,1-8 (2000) respond reversely, in other words, they pushed button 5. T. Takahashi, K. Sumida, Y. Yanase, and H. Riquimaroux, during S- and not during S+ especially for Proc. Spring Meet. Acoust. Soc. Jpn. ,465-466 (2001)

Sound Quality Evaluation of Construction Machine

S. Hatanoa, T. Hashimotoa, Y. Kimurab and T. Tanakab

aDepartment of Mechanical Engineering, Seikei University, 3-3-1 Kichijoji Kitamachi, Musashino-shi, Tokyo, 180-8633, Japan bMechanical Engineering Research Laboratory, Kobe Steel Ltd., 5-5, Takatsukadai 1-chome, Nishi-ku, Kobe, Hyogo, 651-2271, Japan

The evaluation of sound quality of interior and exterior earth-moving machine noise was studied. The evaluation of interior noise was conducted with the video movie recorded simultaneously at the noise recording. The video camera was set in front of the windscreen of the operator’s compartment. For exterior noise, that was conducted with the exterior sight of the machine in operation recorded by the video camera set distant from the machine. The response system was constructed by a personal computer, a liquid crystal projector and a projection screen for collecting time varying subjective evaluation on unpleasantness, powerfulness, sharpness and booming sensation. The results show that impression with moving pictures except for booming sensation was moderate compared with those obtained without pictures. The time variations of sound quality parameters such as loudness and sharpness filtered with an appropriate time constant were calculated for discussing the relation between subjective response and objective measures. Finally a model for unpleasantness was constructed through regression analysis.

INTRODUCTION 7 50 r=0.387(sound with scenery) bad 6 45 r=0.381(sound only) ) 5 40 sone

This paper describes the results of the time varying ( 4 35 evaluation on sound quality of noise emitted from a 3 30 loudness hydraulic powered earth-moving machine. From the pleasant unpleasant 2 good 25 1 20 viewpoint of an operator’s comfort and that of the 0 102030405060708090 (s) inhabitants living near the construction site, exterior as (left-front) well as interior noise was used together with video 7 50 r=0.669(sound with scenery) bad movies showing the interior and exterior scenery of the 6 r=0.632(sound only) 45 ) 5 40 sone machine in operation for evaluation. The time 4 35 ( variation of several sound quality parameters such as 3 30 loudness pleasant unpleasant 2 25 loudness, sharpness, roughness, fluctuation strength good 1 20 and booming index[1] were calculated with appropriate 0 102030405060708090 (s) time constant, i.e., 1.7 seconds for better correlation (left-back) 7 r=0.785(sound with scenery) 50 with the subjective responses. A regression model in r=0.795(sound only) bad terms of these parameters was used for estimation of 6 45 ) 5 40 sone unpleasantness. As a result, good correlation was 4 35 ( obtained between the evaluation of unpleasantness and 3 30 loudness

pleasant unpleasant 2 25 good the estimation obtained by the model. 1 20 0 102030405060708090 (s) (right-front) 7 50 r=-0.036(sound with scenery) bad EXPERIMENT 6 45 r=-0.046(sound only) 5 40 ) sone The machine repeats, “dig-turn-throw-turn” four times 4 35 ( 3 30 loudness

and exterior noise at four locations was recorded, i.e., pleasant unpleasant 2 25 good at left front, at left back, at right front and at right back 1 20 0 102030405060708090 and the distance from the machine to the four locations (s) was 10 m. Interior noise was recorded with the (right-back) total impression for sound with scenery total impression for sound only conditions where an air conditioner switched on and sound with scenery sound only loudness off with interior video scenery. The sound stimuli were presented to the subject through an electro-static FIGURE 1. Evaluation of exterior noise on un- headphone inside the sound proof room. The subject pleasantness and time variation of loudness

was asked to put his/her instantaneous response on obtained utilizing sharpness and the slope of sharpness unpleasantness, powerfulness, sharpness and booming between 8 seconds as explanatory variables and the sensation into a personal computer using a keyboard as one for exterior noise only was obtained utilizing an input device for about 80 seconds from the sharpness and the slope of loudness between 10 beginning to the end. After the evaluation on seconds as shown in Figure 3. Models for unpleasantness subjects were asked to answer their unpleasantness for interior noise were obtained overall impression on sound. 14 males and 1 female utilizing loudness and booming index as explanatory for unpleasantness and powerfulness and 22 males and variables, but due to the shortage of space the results 1 female for sharpness and booming sensation joined were not shown here. as subjects. They all had normal hearing and were aged between 22 to 54 years. Their individual CONCLUSIONS responses in each four factors were averaged over total subjects to get the final results. 1. The results obtained by presenting sound only were more unpleasant, more powerful and sharper than RESULTS those obtained by presenting sound with scenery.

The results on unpleasantness at the four exterior 2. The evaluation of sound quality of exterior earth- locations were shown with the variation of loudness in moving machine noise was influenced by its Figure 1. The responses varied with time according to loudness and sharpness and the one of interior the variation of loudness and sharpness. The results noise was influenced by its booming index and obtained by presenting sound only were more loudness. unpleasant, more powerful and sharper than those obtained by presenting sound with scenery. About total 7 6 impression, evaluation at the right back was better than 5 the one obtained at the right front though loudness at 4 3 2

the right back was larger than that at the right front. pleasant unpleasant 1 This result was due to the fact that large variation of 0 102030405060708090 (s) loudness influenced the total impression. (left-front) 7 6 REGRESSION MODELS 5 4 3 2

We calculated objective measures such as loudness, pleasant unpleasant 1 sharpness and booming index with a time constant of 0 102030405060708090 (s) 1.7 seconds. These were used in the regression (left-back) 7 analysis for rating unpleasantness. The model for 6 unpleasantness for exterior noise with scenery was 5 4 7 50 3 bad 2 6 45 pleasant unpleasant 1 5 40 0 102030405060708090 (s) 4 r=0.401(sound with scenery) 35 r=0.372(sound only) (right-front) 3 30 7 booming index

pleasant unpleasant 6 2 good 25 5 1 20 4 0 102030405060708090 (s) 3 (with air condisioner) 2 pleasant unpleasant r=0.435(sound with scenery) 1 7 50 r=0.414(sound only) bad 0 102030405060708090 6 45 (s) (right-back) 5 40 R=0.900 subjective evaluation regression estimate 4 35 3 30

booming index FIGURE 3. Evaluation of exterior noise on pleasant unpleasant 2 25 good 1 20 unpleasantness and the estimation by a regression 0 102030405060708090 (s) model in terms of sharpness and slope of loudness in 図 車内音不快感評価と音圧レベル(without air condisioner) 10 seconds in case of sound only total impression for sound with scenery total impression for sound only sound with scenery sound only loudness REFERENCE FIGURE 2. Evaluation of interior noise on unpleasantness and the variation of booming index 1. S. Hatano and T. Hashimoto., Proceedindings Inter- Noise 2000, No.233, Nice, 2000, pp. 1-4

Aspects on three methods for paired comparison listening tests A-C. Johansson, P. Hammer and E. Nilsson Division of Engineering Acoustics, LTH, Lund University, Box 118, 223 63 Lund, Sweden

In many acoustic environments there is a need to rank and classify sounds. A frequently used procedure is paired comparison tests. There are, however, a number of ways to perform and analyse this test [3]. In several research projects ties are not allowed or ignored and information might therefore be lost. A comparison of three different approaches is made; no ties allowed, ties allowed but ignored, and ties allowed and used in the analysis.

INTRODUCTION where ai j is the number of times i was selected and a ji is the number of times j was selected (ai j + a ji = ni j). It is sometimes hard for a subject to use and for an By maximizing the natural logarithm of L and using the t π π 2 experimenter to analyse results using scales in the evalu- constraint ∑i i = 1, estimates of i are obtained. ation. There are uncertainties whether the subjects have If the objective of the listening test is to nd out if all used and understood the scale equally. The problem is treatments (or sounds) are perceived similar or not, hy- avoided using paired comparison tests where subjects pothesis tests presented in [1, 2] can be used. If the ob- are asked to judge which of two treatments has a cer- jective, on the other hand, is to receive a ranking of the tain attribute (e.g. a pleasant sound). This paper gives treatments and to say that one of the treatments is better some aspects on some of these models. Two major than another treatment on a certain signicance level, 3 a pair comparison methods exist today, the so-called Thur- different approach is needed. When choosing an appro- stone model [9] and the Bradley-Terry model (BT-model) priate approach it is important to notice that the estimates [1, 2]. They give similar results but since there exist more of πi are not independent and that the variances of each (comprehensive) extensions to the latter, the focus in this treatment are not necessarily homogeneous. It is there- paper will be set on the BT-model and its extensions on fore not appropriate to use ANOVA-tests. We can instead ties [4, 8]. calculate the variance and covariances for the treatments as described in [2]. They form an ellipsoidal region4 as METHODS shown in Figure 1 and its area is dependent on the chosen signicance level. The region tells us that the estimates Test without ties of πi and π j should exist somewhere within that area. A hypothesis test of any point outside the region will be re- The BT-model is a model that gives maximum- jected on the chosen signicance level, and if the region likelihood estimates of the treatment ratings, T ,i = i is crossed by the plane of symmetry the hypothesis that 1,...,t. The probability of choosing T when compared to i they are similar cannot be rejected. For the two differ- T is given in eq. (1) where π ,...,π represent relative se- j i t ent signicance levels in Figure 1 it would be correct to lection properties for the t treatments, i = j,i, j = 1,...,t.1 Z 6 say the treatments are different for the smaller area with π ∞ i 2 the higher α, but not for the larger area with the lower P(Ti Tj)= π π = sech (y/2)dy (1) → i + j (lnπi lnπ j) signicance level. − − The probability can be described as an integral as shown above, where the probability is seen to be dependent on 2 the natural logarithm of the π-values of the treatments. Problems may arise when one or more of the treatments always are chosen in favour of the others. This is the reason why comparison of the treatments 3 The signicance level, α is the probability to reject a hypothesis even should be made on the natural logarithm of the preference though it is true. Lowering α sets a stronger demand on the hypothesis. values, called the true merits. Using a generalization of a 4 By looking at the region we can check if our estimates are indepen- binomial model and its distribution, the complete likeli- dent and the variances homogeneous. If the region forms a circle, the variance is equal for the estimates and the estimates are independent. hood function, L, becomes If an ellipse, nonparallel to any of the axis, is given, as in Figure 1, t ai the estimates are not independent. Furthermore, if it is not parallel to ni j ∏ π L = ∏ i=1 i (2) the symmetric plane, the variances of the estimates are inhomogeneous. a ∏ (π + π )ni j i< j ’ i j“ i< j i j Plots similar to Figure 1 were given when a listening test was performed at the division on various drum sounds from oor coverings. The use of ANOVA-tests would therefore not be correct. A renement of the procedure can be made using contrasts [7], but this procedure is not 1 In all BT- models, comparisons Ti Ti are not allowed. included in this paper. − 1 represented on a linear scale. The probability for prefer- ence of i when presented with j then becomes ν π π πi √ i j P(Ti Tj)= ;P(Ti=Tj)= → πi+π j+ν√πiπ j πi+π j+ν√πiπ j The estimates in the two models and their variances treatment j and covariances are obtained as in the BT-model, us- ing maximum likelihood functions, and are described in their articles. It has been noticed [2, 4] that the models 0 are asymptotically equal and the choice of method is a 0 1 treatment i matter of which idea seems more appealing. David [3] FIGURE 1. Condence regions for two α, comparing treatment points out that the fullment of the choice axiom might i and j. The smaller area is given by a higher value of α. not be required. A difference is, however, that the model by Davidson gives a ranking that is always consistent to a ranking common in sports where a win is awarded 2 Test with ties but not used in the analysis points and a tie 1 point, which the RK-model is not.

This type of test was introduced when no model for REMARKS handling ties existed. The random answers for compar- isons where subjects cannot make a difference are elimi- When allowing ties, subjects might declare a tie al- nated most simply by allowing them in the test but ignor- though they, with some effort, could detect a difference. ing them in the analysis. Some experimenters divide the An investigation [5] on this problem gave the following ties by splitting them equally on the treatments. In [5] recommendations. When discrimination is the objective a test made by Hemelrijk is described where it is proved it is better to prohibit ties as the subjects' efciency of that leaving ties out of consideration makes a more pow- decision might be offset, but when preference is the ob- erful test than if the ties are equally distributed. Today, jective, ties should be allowed as they add information. It however, models capable of handling ties are available, has been noticed by the present author that subject prefer- and this procedure is therefore not recommended, as in- ences are not always normally distributed. Some subjects formation is lost. prefer a darker drum sound while others prefer a higher pitched sound. If ties are prohibited a 50-50 relationship of two treatments would indicate equal treatments, but if Test and analysis with ties ties are allowed, keeping the same relationship and no ties is reported, we can suspect the treatments are not equal The idea behind the Rao and Kupper model (RK- but only given the same amount of preferences. model) [8] is that when the difference between two treat- ments is smaller than a certain value, or threshold, the subjects will declare a tie. The probability of choosing Ti REFERENCES when compared to Tj is therefore (cf. eq.(1)) set to Z 1. R. A. Bradley and M. E. Terry, Biometrika 39, 324-345 ∞ π 2 i (1952). P(Ti Tj)= sech (y/2)dy=π θπ (3) → (lnπi lnπ j)+η i + j − − 2. R. A. Bradley, Handbook of Statistics 4, 299-326 (1984). where η=ln(θ) is the sensory threshold for the subject. 3. H. A. David, The method of paired comparisons, 2nd Ed., The probabilities for preference of j or for a tie is calcu- London: Grifn, 1988. lated using the integral and are given in [8]. 4. R. R. Davidson, J. Amer. Stat. Assoc. 65, 317-328, (1970). Davidson presented another extension to handle ties 5. N.T. Gridgeman, Biometrics 15, 382-388, (1959). based on the BT-model. His approach was to ensure that 6. R.D Luce, Individual Choice Behavior New York:Wiley, an axiom of choice presented in [6] is fullled. The 1959. idea is that alternatives which should be irrelevant to the 7. D. C. Montgomery, Design and Analysis of experiments 5th choice are in fact irrelevant, stated as P(i i, j)/P( j i, j) = Ed., New York: John Wiley and Sons, 2001. π /π , which the RK-model does not full.| It is assumed| i j 8. P. V. Rao and L. L. Kupper, J. Amer. Stat. Assoc. 62, 194- ν ν that P(0 i, j) = P(i i, j)P( j i, j), where is seen as an 204, (1967). Corrigenda, 63 1550. index of| discrimination.| The assumption| of a geometric p 9. L. L. Thurstone, Amer. J. Psychol. 38, 368-389, (1927). mean are based on the fact that the merits, ln(π), can be Product Acoustics: Designing the Sound Qualities of a Manufactured World

L. Fuks

Escola de Música, Universidade do Brasil/UFRJ Rua do Passeio 98, Rio de Janeiro, 20021-290, Brazil

Most objects, devices and systems produce acoustical signals when functioning or when interacting with other objects. Whenever these "by-product" sounds are considered as undesirable, the obvious procedure by acousticians is to employ methods for maximally suppressing them. However, those sounds may in many cases be useful to provide information about the proper behavior of the system. Examples are the sounds produced by car and motorcycle engines, "clicking" sounds from photo cameras and safety locks, "crispy" sounds of cream crackers, "buzzing"sounds of a microwave apparatus, and so many others. These sounds, more than a mere monitor of electro-mechanical performance, may give some kind of aesthetical satisfaction and also convey or suggest information on the qualities of the object and its materials, such as robustness, lightness, "nobility", durability, value, freshness, etc. In order to model, design and modify those object sounds, a multidisciplinary approach is required, involving topics from music perception, music acoustics, psychoacoustics and vibration, science of materials, industrial design, among others. We propose the creation of a new discipline, Product Acoustics, describing some applications, and drawing some guidelines for its establishment.

INTRODUCTION arise: (i) how the chain of mechanisms and interactions result in a particular click?; (ii) how does a good It seems appropriate to bring out this discussion in camera sound like?; (iii) how can the sound be Italy, where Marinetti founded the Futurist movement modelled, improved and possibly applied to a "bad around 1909. Luigi Russolo launched "L'arte dei sounding" camera?; (iv) can the click be muffled or Rumori" futurist manifesto in 1913, proposing a new suppressed when required?; (v) should a digital camera attitude towards the sonic ambient brought by click the classic click?; (vi) should standards be technology. Russolo and co-workers created a new established for product sounds? musical style, a notation system and a set of twenty- To our knowledge, no present discipline offers seven instruments, the “noise intoners” (intonarumori). resources for dealing with such problems, in spite of The sounds and noises produced by machines and most of them being related to acoustics and other products are frequently regarded as undesirable, psychoacoustics. Based on the issues above, we will annoying or irrelevant. Yet, one could hardly think of draft a systematic approach towards a proposed "mute" devices working into complete silence. The discipline, Product Acoustics. “voices” of products convey rich and useful information [1]. VIBRO-ACOUSTICAL ANALYSIS Nevertheless, the sonic outcome of the product is often an uncontrolled, random and overlooked There is a need of a thorough analysis of the dimension in industrial design and engineering. From materials employed, the types of mechanisms, the user's perspective, interaction with the product accelerations and forces involved, the friction and includes the auditory stimuli produced, which may shocks that take place and the way how the vibrations provide clues for the identification and control of its are converted into sounds. This approach combines function and even serve as a source of aesthetic delight. computational solid , acoustics and Furthermore, the sound from an object is in itself an vibration, dynamics of mechanisms, science of object [2]. Thus, it seems valid to gather knowledge on materials, among others. This aspect will considerably the nature and on the aesthetical impact of these benefit from knowledge on the acoustics of percussion stimuli. instruments (idiophones). Let's focus on a photo camera, which is expected to produce the classic click sound whenever a picture is AUDITORY SEMIOLOGY taken. The qualities of this click may indicate whether it is a professional or just a cheap disposable model. Users seem to associate product qualities to a huge However, in situations requiring silence, the camera databank of memorized sounds, which may result in the turns into a disturbing noise source. Several issues impression of robustness, lightness, "nobility", durability, etc. On the other hand, the sounds may phones. Makers aggregate fragments of musical works denounce malfunction, lack of adjustment, the presence and various patterns, apparently without criteria or of cracks in the structure, among others. Therefore, an compliance with any social rules or industrial standards extense library of sounds must be acquired and evaluated by subjects through systematic STANDARDIZATION psychoacoustic and objective procedures. As proposed above, product acoustics is directly related to sound and noise emission, to the SOUND RECORDING, ANALYSIS, quality of the product and to sound communication and REPRESENTATION, MODELLING signalling. There are several ISO standards that refer to noise emission but they are mostly based on maximal admitted loudness levels. ISO-9000 series refers to The sound signals produced by products may be total quality that contemplates several aspects of roughly divided into transient, chaotic and periodic. product acoustics, such as interface with the user, The click of the camera is an example of a transient subjective aesthetical characteristics and convenience sound [see 3]. Each type requires appropriate methods of use. However, to our knowledge, no standards refer for recording, analysis and representation [see 4]. In to sounds produced by equipment that are not designed this aspect, a multidisciplinary approach combining the for sound purposes. We could preliminarly suggest that areas of signal processing, music acoustics, chaos the discipline of product acoustics will open up a whole theory, among others, is required. series of new standards, with clear industrial and project applications. AURAL COMMUNICATION/ ETHOS CONCLUSION Similarly to visual comunication, aural Product acoustics refers to a number of communication impregnates the product with ambience present needs and problems which, however, do not and useful language. Here the sounds are "telling" that correspond to an existing discipline or professional. the camera is about to take the picture, that the picture Therefore, it is likely that a systematic approach in the has been taken, that the battery charge is low, etc. For foundation of this discipline will result in new trends in the user's convenience, these signs might be customized educational, research and technological activities. This to be pleasant and more easily recognized, particularly paper is a first attempt to share the ideas and instigate in a very rich "soundscape" typical of present times [5]. discussion among acousticians. Also, these functions must comply with behavioural constraints (ethos), so that it does not disturb other AKNOWLEDGEMENTS functions and individuals. The user should be able to turn the sound functions off and to reduce the output level. Besides, sounding functions represent a relevant The author profited from fruitful and factor in power consumption of batteries. stimulating discussions with R. Murray Schafer, creator of the World Soundscape Project, to whom this paper is dedicated. SOUND MODIFICATION/ MIMICKING REFERENCES The designer should now be able to intervene on the click sounds. This may consist of modifications in the 1. Dandrel, Louis. The Voice of Things, in Industrial mechanism, materials, superficial texture and Design, of the Century, Ed. by Jocelyn dimensions so that the sounds will be improved to meet de Noblet, Flammarion/PCI, Paris, 1993. some expected results. Mechanical and electronic 2. Schaeffer, P. Traité des objets musicaux. Paris, ressonators and filters may be incorporated to alter the Seuil, 1966. original sound. Experience from the area of sound 3. Gordon, J W. Perception of attack transients in design will be called for in this aspect [6]. In order to musical tones, PhD Thesis, Stanford University, mimic the sound of a desired click, a sound generator 1984. using digital synthesis may be added. This would be the 4. De Poli, Piccialli et Roads (Eds.) Representations case of sounds aggregated to a digital camera which, in of Music Signals. Boston, MIT Press, 1991. principle, could work in complete silence. A common 5. Schafer, R. M. The new soundscape. Vienna: complain to silent cameras is that the photographer is Universal Edition, 1969. not always sure if the picture has been taken. 6. Cogan, R., Escot, P. Sonic design: The nature of The most common, and frequently disturbing, case of sound and music. Englewood Cliffs, NJ: Prentice- sound mimicking is that of the "ring" of cellular Hall, 1976. Evaluation of Reaction to Noise and Vibration – A Survey of Comfort in Airplanes

B. Schulte-Fortkamp1, J. Quehl2, V. Mellert3, H. Remmers4

1,3 University of Oldenburg, Department of Physics and Acoustics, D-26111 Oldenburg, Germany 2 University of Oldenburg, Department of Psychology, Institute for Research into Man-Environment-Relations, D-26111 Oldenburg, Germany 4 Institute for Technical and Applied Physics (ITAP) GmbH, Oldenburg, Germany [email protected]

When sound and vibration are judged concerning comfort, various dimensions structuring this procedure have to be taken under consideration. Necessarily, since the subjective judgments will be influenced by different moderators, the methods have to be adapted to the objectives under physical, psycho-, socio acoustical, and psychological aspects. The survey focusing on perception of sound and vibration was conducted with about 600 subjects from different European countries regarding to flight situations in jets- and propeller airplanes and helicopters. The aim was to develop a comfort index concerning flight situations. The evaluation process on combined effects of sound and vibration integrating interdisciplinary concepts and results will be presented. The work has been supported by the BRITE EURAM Project BE97-4056 “IDEA PACI”

INTRODUCTION Table 1. Semantic Differential for Airplanes and Helicopter The survey focusing on perception of sound and vibration was conducted with about 600 subjects from ITEMS OF THE SD Heli- Airplane different European countries regarding to flight copter bearable unbearable X X situations in jets- and propeller airplanes and comfortable uncomfortable X X helicopters. The aim was to improve the comfort of threatening harmless X X aircraft passengers by the modification of those psycho shaking calm X X acoustic and vibration parameters that physically vibrating not vibrating X X dangerous safe X - correspond to the dimensions distinguishing combined pleasant unpleasant X - acoustic and vibration in aircraft. oppressing liberating X - well-sounding ugly-sounding X - EVALUATION crumpled smooth X - rotating still X - strong weak X - Following a literature study an adequate shrill dull X - methodological instrument for the evaluation of palpable impalpable X - acoustic and vibration experiences in aircraft did not pushy reserved X - muffled not muffled - X exist, different field and laboratory pretests with an acceptable unacceptable - X expert group as well with naive test persons have been regular irregular - X carried out.[1,2,3,4] monotonous varied - X The aim was to develop a context orientated semantic high-frequency low-frequency - X loud quiet X X differential (SD) (Table 1) concerning jet- and rough not rough X X propeller airplanes on the one hand and helicopter on tonal not tonal X X the other.[5] unsteady steady X X sharp not sharp X X 20 15 MAIN TESTS

An aircraft simulation test was carried out in a mock- up (laboratory equipment) with a sample of 117 RESULTS subjects (37 female and 80 male, aged 19 to 61). The helicopter test series have been carried out with 25 With the aid of principal component analysis (PCA), it subjects (13 female and 12 male) taking part at the real was attempted to extract from the SD data independent helicopter flights and 107 subjects (45 female and 62 perceptual dimensions describing the combined male) at the helicopter simulation tests in the mock-up, acoustic and vibration perception in aircraft. [6,7,8,9] 25 of them took part before in the real flight test. Jets and Props incorporation of traditional psycho acoustic parameters, the validation by a thorough design procedure, and the The factor explaining one third of the variance was a availability in three major : German, English, comfort factor related to aircraft interior sound and and Italian. vibration. Comfort seemed to be the counterpart of Following the results of the psychological and psycho specific sound characteristics such as the perceived acoustical research acoustical comfort in airplanes is loudness or roughness as well as particular vibration everything contributing to the well-being and it attributes (e.g. "vibrating"). The second dimension was constitutes an improvement of given conditions in an associated with time characteristics of the flight airplane. Acoustical comfort contributes to a general situations like "monotonous" and "regular" and further comfort definition by parameters which usually describe comfort, but as counterparts to the comfort

1 parameters special acoustical parameters play a significant role.

0,8 unsteady ACKNOWLEDGEMENT

0,6 shaking BRITE-EURAM project "IDEA PACI" / BE97-4056

0,4 vibrating threatening REFERENCES sharp 0,2 rough 1. Janke, W. & Debus, G., Die Eigenschaftswörterliste: comfortable EWL; eine mehrdimensionale Methode zur 0 loud -1muffled -0,5 0 0,5 1 Beschreibung des Befindens. Hogrefe, Göttingen, 1981. acceptable bearable -0,2 2. Osgood, C.E, Suci, G.J. & Tannenbaum, P.H., The measurement of meaning. University Press of Illinois,

-0,4 Urbana, 1957.

3. Osgood, C.E., Focus on meaning. Mouton, The Hague, -0,6 1976.

regular -0,8 4. Pineau, C., The psychological meaning of comfort. monotonous International Review of Applied Psychology, 31, pp. 271-283, 1982. -1 Factor 1 [37.3%] 5. Quehl, J., Schick, A., Mellert, V., Schulte-Fortkamp, B., Figure 1. PCA for all propeller and jet flight situations Remmers, H., Effects of helicopter and aircraft interior noise and vibration an passengers’ comfort sensation vibration qualities like "unsteady" and "shaking". The and subjective well-being. J. Acoust. Soc. Am., 105 (2) third factor was confined to the perception of tonality. (1084) and ACUSTICA/acta acustica, 85, p. 158 ( 1999).

6. Quehl, J., Schick, A., Mellert, V., Schulte-Fortkamp, B., Helicopter Remmers, H., Evaluation of combined aircraft interior sound and vibration effects on passengers` well-being By PCA of 5 flight situations 3 factors were detected and comfort sensation: the elaboration of a concept- describing the combined acoustic and vibration specific methodologoical instrument. Results of the 8th perception. The most important dimension explaining Oldenburg symposium on psychological acoustics. bis, one third of the variance was a comfort factor related to Oldenburg, 2000. specific noise characteristics such as loudness or roughness, vibration, and particular attributes 7. Quehl, J., Schick, A., Mellert, V., Schulte-Fortkamp, B., Remmers, H., Hauptdimensionen einer kombinierten describing the sensations. The second dimension was Geräusch- und Vibrationswahrnehmung in associated to further vibration qualities like rotating Flugsituationen: Auswertungen zum semantischen and shaking. The third factor was confined to the Differential. Fortschritte der Akustik - DAGA 2000, perception of psycho acoustic parameters. Oldenburg, 2000.

CONCLUSION 8. Remmers, H., Reckhardt, C. & Bellmann, M., A system of natural reproduction of sound and vibration. J. The process conforming the evaluation procedure to Acoust. Soc. Am., Vol. 105 (2), Pt. 2. (1999) the objectives led to two different semantic profiles 9. Zwicker, E., Psychoakustik. Springer, Berlin, 1999. concerning airplanes and helicopters. The major innovations were in: the combination of noise and vibration, the novelty of the semantic attributes, the Low Frequency Perception in Urban . A Cognitive Approach Catherine Guastavinoa, Danièle Duboisb, Jean-Dominique Polacka and Christine Arrasc aLaboratoire d’Acoustique Musicale, Université Paris 6, 11 rue de Lourmel, 75015 Paris, France bLaboratoire Cognitions Pratiques et Ergonomie, 44 rue de l’Amiral Mouchez, 75014 Paris, France cAcouphen, BP 2132, 69603 Villeurbanne Cedex, france

An investigation of the subjective impression of low frequencies due to traffic noise in urban soundscapes was carried out, using open questionnaires in three different contexts to better understand how low frequencies affect people outdoors. A preliminary survey was sent by mail. People were then interviewed in actual outdoor environments, which were recorded simultaneously. The recordings were used for listening tests in an acoustically damped room. Presented in this paper are the main results that can be drawn from the psychological exploration of cognitive categories related to low frequency phenomena and their representations in language. The comparison of the results obtained in the different contexts sketches some theoretical and methodological issues.

OVERVIEW tween individual representations in language (in our case, linguistic devices involved in the description of low fre- In urban areas, noise stems from a wide variety of quency phenomena) and shared conceptual cognitive rep- sources, many of which contain predominantly low fre- resentations. quencies. These frequencies easily propagate over large distances, are able to proceed around obstacles and fill the urban space completely. Complex multisensory pro- RESULTS cesses are involved in the perception of the low frequency range (below 200 Hz) where audible sounds become tac- Presented below are the main results that can be drawn tile vibrations. The increasing problem of noise annoy- from the psychological exploration of cognitive cate- ance reveals the limits of physical description of noise to gories related to low frequency phenomena and their rep- measure the subjective impression, and suggests a more resentations in language. cognitive approach to noises as meaningful events that af- The linguistic analysis conducted on our verbal data fect people. shows that the low frequency phenomena are perceived People living in three French cities (Paris, Lyon, through the following two semantic categories : “sound Nantes) were questioned about their appraisal of urban events” that can be attributed to an identified source, and soundscapes and their descriptions in three different con- “ambient noise of the city” or “background noise” (’bruit texts. A preliminary survey was sent by mail to 80 people. de la ville, bruit de fond’ in French) where no specific Answers refer to memorized representations of familiar event could be isolated. urban soundscapes. A shorter oral version was used for A large variety of linguistic devices were observed in 42 interviews in real outdoors environments, where all the description of ambient noise. Subjects found it very are involved. Recordings were carried out simul- difficult to identify and describe. They described it in a taneously. The soundscapes were selected from a list of very global manner, as a whole rather than sources emit- locations identified as representative of city noises (Paris) ting noise, and expressed it in terms of effects perceived [1, 2]. The recordings were used for listening tests with by the subject but also physical properties of the sig- no visual information, in an acoustically damped room, nal. The physical descriptions refer to the timbre (’muf- using a stereo set-up (Studer A1) and a (JBL fled, muted’), the temporal structure (’continuous, per- 4645 C) below 100 Hz. 29 subjects listened to 6 different manent’) and the envelopment (’it is all around me’). Re- urban soundscapes. garding qualitative evaluation, the ambient noise is cate- In all three cases, semi-structured (’semi-directifs’) gorized as non negative (’not unpleasant’) for most peo- questionnaires were designed with the same open ques- ple, and even comforting for some. It is considered as tions using very general terms (’feel, be affected’) in or- a sign of human activity, characteristic of city life and der not to influence the judgment or confine the answers therefore well accepted. in predefined categories. The present study relies on the Sound events are described in terms of sources (’en- analysis of psycholinguistic processes that mediate be- gine, trucks’) and action or movement of the source gen- erating the noise (’the engine idles’). Subjects described room has been redesigned so as to remove the visual ref- how they were affected by the sound events using dever- erence of the , masking them from the test bal adjectives, i.e. adjectives based on a verb, which refer subject for future experiments. mainly to a hedonic scale (’annoying’). Their judgment The comparison of the data obtained in the three con- of the acoustic phenomena is closely linked to their ap- texts shows similar results as far the sound events are con- praisal of the source itself and what is semantically as- cerned. This confirms that the stereo set-up used in previ- sociated with it. Few descriptions of the sound event is ous studies [1, 2], along with the instructions given to the given in terms of physical parameters (34% of the occur- subjects is ecologically valid in terms of source identifi- rences refer to physical properties whereas 67% describe cation. However, regarding ambient noise, the results are effects on the subject1). A sound event is something that quite different. Ambient noise was mostly described in occurs and can be heard during a particular interval of terms of physical parameters (59% of the occurrences vs. time, thus not abstractable from time and space. The tem- 25% in situ) and not in terms of effects (14% of the occur- porality is implicitly delimited in the sound event, which rences vs. 56% in situ) and always processed as a sound. may explain why no description of the temporal structure Moreover, subjects complained that they did not feel as if is given by the subjects. they were "there" and that the envelopment provided by Recent psycholinguistic studies show that acoustic the subwoofer was not consistent with the frontal image phenomena can either be processed as noises, perceived of the other speakers. as effects of the world on the subject, or in a more an- These results show that the same acoustic phe- alytic manner as sounds, perceived as objects of the ob- nomenon could give rise to two different cognitive ob- jective world [3]. Results show that the sound events are jects, namely a noise or a sound, that integrate proper- clearly processed as noises, whereas the ambient noise ties of mental representations into physical descriptions is more abstracted from the object-source (since the pro- of the stimuli. This theoretical point should be taken into cess of source identification fails) and more frequently account in methodology for listening tests since ecolog- described as a sound by means of its physical properties, ical validity depends on the purpose of the study. Steps but on the other hand, it is also processed as a meaningful were taken to overcome the limitation of stereo reproduc- phenomenon referring to the presence of human activity. tion by using multi-channel reproduction in future exper- iments to improve immersion, which turns out to con- tribute to the cognitive representation of ambient noise. Nevertheless the difference between listening tests and Ecological validity everyday listening situations may also be imputed to the artificial laboratory conditions and the required processes Some methodological difficulties were encountered of abstraction. with the listening tests. Recording of six different sound- scapes were presented, each 5 minute long. Instructions were given to direct the subjects’ response strategy to- ACKNOWLEDGMENTS wards an everyday listening situation, so that they react, to some extent, as if they were in a actual situation, ac- This research is financed by the French Ministry of cording to the concept of ecological validity developed by Environment. Gibson [4]. For each soundscape, subjects were asked to spend a few minutes acclimating themselves to the recre- ated acoustic experience, and then answer open ques- REFERENCES tions. The recordings were reproduced on a 2.1 format (2 speakers and a subwoofer, 1.2 meter away from the lis- 1. Maffiolo, V., Caractérisation sémantique et acoustique de tening spot) in an acoustically damped room. The use of la qualité sonore de l’environnement sonore urbain. Thèse the subwoofer was one factor of the test, resulting in only Université du Maine, Le Mans (1999). half of the test examples using it. A total of 29 subjects 2. Vogel, C., Etude des signaux sonores d’avertissement. participated in the experiment. Thèse Université Paris 6 (1999). Most subjects were impressed by the low frequency 3. Dubois, D., Categories as acts of meaning : the case of recreation even when the subwoofer was not being used. categories in olfaction and audition, Cognitive Science The conclusion is that the visual setting affected their im- Quaterly, 1, pp. 35-68. (2000). pression of low frequency. Subsequently, the listening 4. Gibson, J., The Ecological Approach to Visual Percep- tion, Lawrence Erlbaum Asssociates, Hilldale, New Jer- sey (1979). 1 One answer may give rise to several occurrences Detection threshold of a periodic phase shift in music sound R. Nishimura, M. Suzuki and Y. Suzuki Research Institute of Electrical Communication, Tohoku University, 2-1-1 Katahira, Aoba-ku, Sendai, Japan

Ability of the human auditory system to detect periodical phase change in musical sound was examined. Results of experiments based on the A-X-B paradigm showed that the human auditory system cannot detect musical sound change if the cycle period of phase change is lower than about a few hundred Hz.

INTRODUCTION 0

The human auditory system has been believed to be

insensitive to sound phase. For example, it is difficult rad) −0.5 π for human beings to judge change in timbre by phase dif- × ference in high frequency components of complex tones [1]. In low frequency ranges, however, timbre of complex −1 tones clearly depends on phase difference between com- ponents [2]. Accordingly, belief that the human auditory −1.5 system is entirely insensitive to sound phase is inaccu- Phase Modulation ( rate. Hence, insensitivity of human auditory system was −2 measured in terms of ability to detect periodical phase 0 0.2 0.4 0.6 0.8 1 rotation in musical sounds. Phase rotation was artificially Normalized Frequency generated by means of all-path filters. FIGURE 1. Phase characteristics of all-path filters with differ- ent ω0s. PHASE MODULATION BY ALL-PATH FILTER 0

An all-path filter is expressed in the s-plane as follows: rad) −0.5 π × ω 2 ¤ 0 ω2 s s ¥ 0 −1

Q ¦ H s ¡£¢ ω (1)

2 0 2

¥ ¥ ω s s 0 Q −1.5

Phase Modulation ( Norm. Freq.= 0.1 Norm. Freq. = 0.6 where Q and ω0 are parameters determining filter phase −2 characteristics. Figure 1 depicts the phase characteristics 0 200 400 600 800 1000 of all-path filters with different ω0’s. Phase shift of ¤ π is Sample realized at frequencies corresponding to ω . By changing 0 FIGURE 2. Phase rotation at two frequency components. Q or ω0 as a sinusoidal function, filter characteristics vary periodically. This is referred to hereafter as “phase rota- tion”, and frequency of the periodical change in phase as “rotation frequency” . Figure 2 shows phase rotation at represented phase modulated sound. Either A or B is the two frequencies. same as X, and the other is the original. Four kinds of sound signals were used in these exper- iments. They were: an instrumental music selection, a EXPERIMENTS song by a female singer, pink noise, and a pulse train of 1ms width with 2ms cycle. Parameter Q in the Eq. The minimum frequency to detect phase rotation was (1) was set to one and two. Parameter ω0 was changed measured through a listening experiment. The A-X-B periodically within a range of 8 kHz to 20 kHz. Sound paradigm was employed to measure ability to distinguish pressure level of each stimulus differed slightly depend- phase modulated sound from the original one. X always ing on the sound type, from 74 dB(SPL) to 77 dB(SPL). Five subjects, four males and one female, took part in the experiments. All were in their 20’s with normal hearing [instrumental music ] acuity. 100

RESULTS AND DISCUSSION 75

50 Figure 3 shows results of experiments. Ordinates of these figures exhibit the rate of correct responses. The au-

Correct answers (%) 25 Q=1 thors regard the rotation frequency at which 75% correct Q=2 judgments are realized as the detection threshold. The 0 1 2 3 10 10 10 detection threshold for instrumental music was about 300 Rotation Frequency (Hz) Hz. It was about 100 to 250 Hz for the song by a female

singer, which is lower than that for instrumental music. [song by a female singer ] Pink noise exhibited the highest detection threshold of 100 about 4 kHz under Q ¢ 1 condition among all stimuli ex- amined. The pulse train, on the other hand, exhibited the lowest detection threshold. 75

Detection thresholds under Q ¢ 2 condition were gen- 50

erally higher than those under Q ¢ 1 condition. This may be due to the small frequency region where phase shift

Correct answers (%) 25 Q=1 occurs becoming broader along with increased Q value. Q=2

This would also result in decreased cues for detection 0 1 2 3 10 10 10 of phase modulation, particularly in a lower frequency Rotation Frequency (Hz) range.

[pink noise ] CONCLUSION 100

Ability of the human auditory system to detect any de- 75 terioration of sound introduced by artificial phase rota- 50 tion in high frequency range was measured. Results of listening tests revealed that human beings cannot notice

Correct answers (%) 25 Q=1 rotation frequencies under a few hundred Hz. From the Q=2

technological point of view, this knowledge has potential 0 2 3 4 10 10 10 for exploitation in developing an innovative watermark- Rotation Frequency (Hz) ing algorithm for musical sound.

[pulse train ] ACKNOWLEDGMENTS 100

This study was partially supported by a Grant-in-Aid 75 for Development of Innovative Technologies (Millen- 50 nium Project, 12107) by the Ministry of Education, Cul- ture, Sports, Science and Technology.

Correct answers (%) 25 Q=1 Q=2

0 0 1 2 10 10 10 REFERENCES Rotation Frequency (Hz)

1. K. Ozawa, Y. Suzuki, and T. Sone, Monaural phase effects on timbre of two-tone signals, J. Acoust. Soc. Am., 93(2), FIGURE 3. The rate of correct responses at each rotation fre- 1007-1011 (1993). quency. Respective panels correspond to instrumental music, a 2. B.C.J. Moore, An Introduction to the Psychology of Hear- song by a female singer, pink noise, and a pulse train. ing, Academic Press Ltd., (1989). An Annoyance Meter for Squeak-and-Rattle Diagnostics D. Dufournet a, P. Susini b, M. Slama b, S. McAdams b

a 01dB-Stell, Limonest, France b IRCAM-CNRS, Paris, France

For sound quality requirements in industrial production, it no longer suffices to qualify acoustic sources in terms of level and frequency. Notions of robustness, safety and product quality are more and more related to auditory sensations, notably as concerns components of automobiles and domestic appliances. This paper presents a new type of device that offers an innovative alternative to the classical sound level meter. Dedicated to the real-time characterization of sound quality, it allows the measurement of quality related to the perceptions of consumers and takes into account the evaluation context. Based on a light and autonomous system, the device offers a palette of psychoacoustic criteria which can be dynamically integrated into a model of annoyance and/or comfort in order to adapt it to the phenomenon under study. An example of a psychoacoustic method is presented in the framework of a national research project on comfort in automobile interiors. The laboratory work based on panels of listeners allows the development of a model that is immediately integrable into the device. The access to a database of downloadable and sharable models is also discussed.

INTRODUCTION PSYCHOACOUSTIC EXPERIMENT The improvement of acoustic comfort in passenger car interiors is currently a major concern of car Psychoacoustic methods were employed to answer manufacturers and suppliers because it corresponds to the following questions: an increasing demand from customers. • What is the minimum acoustic level of an S&R The past effort consisting in improving noise signal that will allow it to be detected over the ambient reduction with the help of absorption and masking driving noise? devices has led today to an increase in driver • How can the annoyance sensation due to a detectable sensitivity to transient sounds, namely “Squeaks and S&R signal be characterised and measured? Rattles” (S&R). These signals, produced by cockpits, It does not seem realistic to evaluate these answers for seats, and accessories, due to vibrations of the vehicle, all driving situations and for each possible S&R signal. are synonymous with low quality and have important Preliminary work conducted in collaboration with car economic consequences. manufacturers and suppliers consisted in identifying The correction of these problems is often realised by life situations of interest and the main S&R signals human experts who decide on their seriousness and the that often appear. Real recordings on the road (life necessary actions to be taken. This subjectivity has situations), completed by test bank recordings using several drawbacks: non-stability of the judgements, no car and component shakers produced a database that on-line production control, no possibility to establish was used for the psychoacoustics experiments. long-term monitoring on subparts of the vehicle. It is The experimental protocol consisted of choosing a thus necessary to establish adapted sound-quality panel of potential customers for listening tests in an metrics and to propose the methods and audiometric chamber, using several mixtures of instrumentation that can measure them in real time. ambiant noise for various life situations and S&R An 18-month collaborative research project, signals. Classical psychoacoustic methods and tests “SQUAD”, financed by the French Ministry of were defined and applied: adjustment and adaptive Research and Technology, brought together French methods (3 down-1 up) for a detection study and automobile and parts manufacturers, psychophysical scaling methods for annoyance psychoacouscians, and noise engineers, to study these assessment. Two series of experiments based on metrics and to propose a prototype system, based on a potential customers and audio experts were conducted, light hardware solution: a Squeak-meter. focusing on detection thresholds and annoyance This paper presents the psychoacoustic methods and ratings for each panel. the main features of the instrument. Extensions and The following figures present some of the results project perspectives are also discussed. obtained in detection and annoyance experiments. ANNOYANCE S&R METER

The integration of previous results in the Symphonie PC-based system was realised in dBQMark (Quality Mark) software. Based on a real-time specific loudness (20 ms window) and other metrics, this software proposes some original functionalities: • real time “in-car” learning of new life situations; • SQUAD real-time metrics (detection and annoyance) allowing an evaluation during travel by moving in front of components to be qualified; • real-time user metrics based on polynomial forms of classical criteria (Loudness, Sharpness, etc..) FIGURE 1. Masked detection thresholds obtained for the customer panel (dotted line) and audio experts (continuous line) for presentation of ten S&R signals in a 130 km/h motorway ambient noise. Vertical lines indicate ±1 standard deviation of the mean. The experts are more sensitive by about 3 dB, but their curve has the same form as that of the customer panel.

FIGURE 3. dBQMark prototype software: real-time production of quality note and S&R annoyance metrics.

CONCLUSION FIGURE 2. Comparison of annoyance ratings The SQUAD project has produced an operational (expressed in absolute units) between experts prototype that will now be validated in real situations (continuous line) and customers (dotted line) for ten by car manufacturers and suppliers. With this S&R signals in a 130km/h motorway situation. Some instrument, an objective quality note can be produced differences appear for some S&R signals: customers for each kind of S&R signal in several life situations. are more annoyed by some sounds than are the experts. Applications include default tracking, long-term monitoring, and systematic on-line characterisation of The results obtained from these psychoacoustic car interiors. Some extensions of the software will be experiments were analysed using both principal realised with a multi-channel acquisition board in components analysis and analysis of variance to order to establish automatic localisation of S&R establish relations with physical features of the defects. acoustic signals. Two algorithms, based on an estimation of specific loudness (ISO 532 B) and various psychoacoustic ACKNOWLEDGMENTS criteria were developed. Strong correlations with human judgements were found (r>0.8 for detection, We thank Visteon, Renault and PSA for participating r>0.85 for annoyance) and are proposed to be in the SQUAD project and for bringing to bear their integrated into the real-time prototype. knowledge of S&R treatment and analysis. Measurement of the Noise Quality of Post-Sorting Machines by Jury Testing

F. Crennaa, B. Ferraria, M. Paneroa, G. B. Rossia, R. Weberb

aDepartment of Mechanics and Machine Design, University of Genoa, Via Opera Pia 15 A, 16145 Genova, Italy bAKUSTIK - FB Physik, Carl von Ossietzky Universitaet Oldenburg, Postfach D-26111, Oldenburg, Germany

Although current measurement of the noise environment in industrial plants is based on simple weighted sound pressure level, there is some growing awareness of the need of more sophisticated methods employing ear-related metrics. This is particularly true for high-technology devices, such as machines for the automatic sorting and addressing of mail items, in which case both producers and users accurately account for ergonomic aspects. So a research has been undertaken in order to construct a scale of pleasantness of typical sounds generated in a wide variety of post-sorting machines, in different environments (testing room and plant) and under different operating conditions. After a complex measurement campaign, a complete database of recorded signals has been constructed, and two kinds of jury tests have been designed, one based on pair comparison forced-choice trials, the other on magnitude estimation with a fixed anchor signal. Results show the feasibility of the approach and give information for a refinement of the testing procedures, in order to get a reliable and robust result.

METHODOLOGICAL PREMISE gradually, a more thorough understanding of the judgement forming process may be pursued. To that Current industrial practice for assessing the noise goal three major steps may be foreseen [8-9]:1 environment in working places is essentially based on Identification of the class of sound-objects under the measurement of weighted sound-pressure levels, as investigation and selection of set of them able to

Legend: FEEDER ROBOT FOR AUTOMATIC TRAYS DOWNLOAD ADDRESS READING WRAP UP MACHINE MANUAL SORTING TRAYS PACKETS DOWNLOAD

MICROPHONE POSITION OPERATOR POSITION

FIGURE 1. Schematic diagram of a post sorting machine required by the related technical norms [1]. Now this represent the whole class. 2 Assessing the quantity is known to be just a rough indicator of measurability of the pleasantness index over an actually perceived loudness [2], its use being appropriate (psychophysic) scale. 3 Defining a historically justified by its simplicity of reference (primary) measurement procedure. Of implementation, especially by analog oriented course, due to the experimental nature of the work instrumentation [3]. At present several factors push several interactions may be needed prior to obtaining toward a more elaborate approach, including the reliable and robust results. The current state for the availability of digital flexible and portable present research will be briefly presented in the instrumentation, the progress in psychoacoustic studies paragraphs to follow. [2,4] and an increasing sensitivity towards ergonomic aspects of workplaces [5-7,11]. In view of ergonomic POST SORTING MACHINES aspects, it seems sensible to look for some indicator accounting for the overall acoustic sensation, leading In figure 1 is presented a general post sorting to a pleasantness/annoyance judgement. At present, no machine configuration. After the manual or automatic universally accepted acoustic pleasantness index is infeed (a) the input post is transformed in a train of available, so that a search for an appropriate one single envelopes in order to be able to read the address should be done for each class of sound objects of (b). Then the envelope runs over a delay line, until the practical or of theoretical interest. In this way both a address image is processed and decoded. Then it goes progress in each specific application field and, into the proper output tray which, when full, can be downloaded manually (c) by an operator, or automatically by a robot (d). In this case post packets clusters, it is not over the set of the signals, since can be wrapped out automatically (e) before clusters have not the same number of signals. downloading (f). The overall length of such a plant 100 Magnitude estimation test varies according to the number of download elements, 90 a common length is around 30m; such a plant can sort 80 70 more than 40000 letters per hour. Measurements have 60 been performed both on operators’ positions and on 50 additional position which may be occasionally 40

30

reached, for instance for surveillance or maintenance mean score reasons. In the research the noise patterns of a 20 representative set of post sorting machines has been 10 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 considered. signal number FIGURE 2. Signals’ mean score from magnitude test ACOUSTIC DATA BASE Preliminary results form both kind of tests have been As mentioned in the premise, the measure we look processed and compared. They present a general for should be defined of a representative subset of the agreement, even if there are discordances for signals considered objects. So the first essential step is the with similar magnitude. Probably there is the effect of construction of the database of reference acoustic some inconsistency in subjective judgements (not signals. This phase has at least two essential points: - removed) and a biasing effect of the clusters. On the define the structure of the database; - actually acquire basis of these first results the investigation will the signal considering practical limitations involving continue according to the following lines: reduction of machine accessibility. Measurement conditions are the set of signals under test by elimination of signals particularly severe if one considers the dimensions and not well characterized; new magnitude estimation test, the complexity of the involved machines. The structure with a modified anchor definition, by randomly of the data base has been defined according to machine selecting single signals; cluster reorganization on the type, working condition and recording position. At base of the results of these first tests. present it includes more than 200 signals. ACKNOWLEDGMENTS TEST DESIGN The authors are grateful to ELSAG Bailey SpA for From the database a selection of a first tentative giving the possibility to perform the experimental reference set have been made. It includes n = 40 work. Special thanks to Dr. Leonardo Roncarolo for signals lasting T = 15 s each. Two kinds of tests have precious advice and encouragement. been implemented: (forced choice) pair comparisons, and magnitude estimation with an anchor signal. A REFERENCES problem at this stage was the high number of trials required. It was faced in two ways: by clustering the 1. ISO 11200:1995. Acoustics – Noise emitted by machinery and equipment – Guidelines for the use of basic standards for the signals into 9 clusters each including signals that after determination of emission sound pressure levels at a work a preliminary listening were considered “similar”. station and at other specified positions. Both tests use a random sorting mechanism, ensuring 2. Zwicker E and Fastl H Psycho-acoustics, Springer Verlag, that each cluster was treated the same number of times 1999 3. Yang S J, Ellison A J Machinery noise measurement, Oxford as the others. This approach allows each individual to University Press, 1985 performs as much trails as he/she likes, without getting 4. Cook P R Music, cognition and computerised sound, MIT tired, whilst the results may be easily cumulated Press, 1999 without biasing the distribution; clustering allows also 5. Meister D The history of human factors and ergonomics, Lawrence Erlbaum Ass, 1999 a quicker convergence on the set of clusters (i. e. on a 6. Salvendy Handbook of human factors and ergonomics Wiley partition of the space of object under investigation). II, 1997 7. Bridger Introduction to ergonomics, Wiley, 1995 EXPERIMENTAL RESULTS 8. Roberts F S 1979 Measurement theory, Addison Wesley, Reading MA The result of the magnitude estimation test is 9. Rossi G B, Crenna F, Belotti V Measurement of perceived noise quality for the ergonomics of post-sorting machines shown in figure 2. It consists of about 500 scores, from IMEKO TC 18 Congress, Sapporo, 2001 a jury of 30 people. The figure presents the mean score 10. Purghé Methods of psycho-physics and uni-dimensional for each signal; bars indicate the mean dispersion. Due scaling (in Italian), Bollati Boringhieri, 1999 to cluster organization, sorting is uniform over 11. Lyon R H Designing for product sound quality, M. Dekker, 2000 Continuous Evaluation of Sound Quality in a Bus

E. Parizeta, L. Segauda, J.R. Kochb and D. Barbelonb

aLaboratoire de Vibrations et d’Acoustique, Insa Lyon, 20, avenue Albert Einstein, 69620 Villeurbanne, France bVibratec, 28 Chemin du Petit Bois, BP 36n 69131 Ecully, France

The main goal of this work is to improve noise comfort in a bus. A sound sequence of long duration has been recorded with a dummy head located in the bus meanwhile the bus was driven in a typical way, including many usual sound events (idle noise, acceleration noise, cruising noise, opening and closing of doors, switching on an off the air conditioning system). This sequence was submitted to listeners who had to continuously evaluate the comfort of the noise. The evaluation was made by the continuous categorical method used by Weber for loudness evaluation. The results allow the identification of the most uncomfortable sound events. Such a procedure makes possible the subjective evaluation of complete and realistic recordings in a driving vehicle.

INTRODUCTION listener had to move the potentiometer in a position he could freely choose along the whole range. Most studies dealing with noise comfort in The test was driven by Matlab software running on a transportation are related to steady state noises, PC computer. An audio card played the sound; a low- though such noises are not so often present in a road frequency generator provided a sine signal, the vehicle, as traffic conditions forces the driver to slow amplitude of which was adjusted by the down or accelerate quite frequently. Some potentiometer; this signal was then digitalised by researchers have developed methods to continuously another audio card. Finally, the envelope of this evaluate sound loudness of traffic noise, as heard by signal was computed, which gave the continuous residents : Kuwano and Namba [1] , who used a evaluation of noise comfort from the listener. categorical method, Weber [2] who used a 48 people participated in the test; they were mostly continuous categorical method. Moreover, Susini and students, did not mention any hearing impairment McAdams [3] asked listeners to evaluate loudness of problem and were paid for their participation. First of sound recorded in cars, using a continuous all, they were presented the whole sequence; then categorical method and a cross-matching one. they made two successive evaluations of it, separated The goal of this study was to adapt one of these by a short questionnaire about the test and the noise. methods to the evaluation of noise comfort, which is a multi-dimensional parameter (as compared to RESULTS loudness). It was applied to noise in a city bus, in order to identify the least comfortable events that Individual Behaviours passengers can heard during a typical journey. First of all, we had to be sure that the task could be MEASUREMENTS AND PROCEDURE achievable by listeners. This can be evaluated from two facts : A dummy head was placed in a bus driven without - each listener gave two continuous evaluations of any passenger on a journey representing usual the same sequence. The correlation coefficient conditions of a bus (engine idle, run up, constant between these two evaluations is always good speed at 50 km/h, coast down, opening and closing of (upper than …); doors, etc…). The overall sequence duration was 162 - the time-averaged value of each listener's s. This sequence was submitted to listeners through evaluations is similar to the overall evaluation of headphones in a quiet room. the sequence given by the listener in the The continuous categorical method, as defined by questionnaire separating the two evaluations. Weber [3], was used. A potentiometer was given to Therefore, it seems that listeners succeeded in the the listener; it indicated five categories (“very task, though about half of them (44 %) found it comfortable”, “comfortable”, “a little bit "rather difficult" or "difficult". uncomfortable”, “uncomfortable” and “very When looking at listeners results, it appears that two uncomfortable”). While hearing the sound, the individual behaviour can be found (figure 1). Relation with loudness

Figure 3 shows the loudness of the sequence, computed according to the ISO 532B procedure, by the MTS software Sound Quality.

70

60

50

40 SoneGD 30

FIGURE 1 : Example of individual answers 20 0 40 80 120 160 Some listeners try to closely follow their evaluation sec. of sound comfort with the potentiometer (bottom curve of figure 1), whereas other ones "integrate" FIGURE 3 : Loudness (ISO 532B) of the sequence their answer on some categories (top curve of figure 1). Such a difference had already been pointed out by From a comparison between figures 2 and 3, it Weber [2]. appears that noise comfort is closely linked to loudness, as it is often the case. But loudness can lead to an overestimation of some events (as doors closing Average Results at 75 s) or an underestimation of other ones (as the starting on of ventilation system at 90 s). The average continuous evaluation is presented in figure 2. CONCLUSION

The method of continuous evaluation, which has already been proven to be useful for loudness studies, can also be used in studies dealing with aesthetic aspects of sounds. It allows to identify, in a complex sequence, the events where comfort is reduced and, therefore, to indicate to engineers where to put efforts on in order to improve noise comfort. This study is now going on by focusing on such events.

FIGURE 2 : Average evaluation ACKOWLEDGEMENTS

From this curve, the least comfortable events clearly The authors are grateful to P. Deprez and M. Salson, appear : from Ratp, who authorised the publication of this - both accelerations are the most prominent ones paper. (they occur in the time intervals [5–20 s] and [95–120 s]. It should be kept in mind the fact, mentioned by Susini and McAdams [3], that REFERENCES listeners may have followed the change of 1. Kuwano S. and Namba S., Psychol. Res. 47, 27-37 tonality, due to the engine run-up, rather than (1985). their appreciation of comfort during the 2. Weber R., "The continuous loudness judgement of accelerations, so that these results must be temporally variable sounds with an "analog" category carefully accepted; procedure", in Proc. Of 5th Oldenburg Symp. On - the opening and closing of the doors (near 65 Psycholog. Ac., 1991, pp. 267-294. and 80 s) are clearly uncomfortable; 3. Susini P. and McAdams S., Acta Acustica, 86, 515- - as is the starting on of the ventilation system 525 (2000). (near 90 s). Product Sound as an Important Part of Product Design C. L. Fog DELTA, Division Acoustics & Vibration, Building 356, Akademivej, DK-2800 Kgs. Lyngby, Denmark Phone +45 45 93 12 11, fax +45 45 93 19 90, e-mail [email protected]

In modern society we are almost constantly surrounded by products, whether we are at home, at work, on vacation, or on our way. We suggest that one essential determinant of “Quality of Life” is the noise or the Product Sound produced by these ubiquitous sound sources. A product that rattles, rumbles, or screeches unpleasantly has a very different effect on the user regarding the perceived va- lue and quality than one with an excellent Product Sound. Product Sound is therefore becoming a still more important part of the pro- duct design just like functionality, form, material, colour, and other major design parameters. For designers and product developers, however, it is often rare to work directly and consciously with the design of the Product Sound. In this paper we will highlight where and how Product Sound can be implemented in the design and product development process. A method using the combination of marketing research, psychoacoustics, and optimal mechanical design will be presented.

INTRODUCTION It should also be endeavoured to determine a target sound – without a goal it is not possible to know in The quality of a product perceived by users and what direction the design work shall take. There are a other observers in the vicinity of the product depends number of tools to be used in this connection: on a number of product attributes such as appearance, S Benchmarking response to user activities, function, noise/sound, - To further develop/improve an existing product weight, smell, taste/flavour, and tactile characteristics. the market can be examined and competitors’ We talk about the sound quality, the visual quality, the products purchased in order to get a comprehen- tactile quality, the quality of user interfaces, etc. [1]. sive view of how far they have come and what Although the Product Sound should not be treated sounds are popular. as an isolated phenomenon, it makes sense in many S cases to optimise this characteristic to improve the Positioning overall perceived quality and thereby the user satis- - Is there a special sound picture forming a kind of faction regarding the product. reference as to perception of quality, cf. door Product Sound can be considered as information, slamming of certain cars? which is relevant in relation to sounds in e.g. the user - Is our product positioned appropriately in relation interfaces of the product/system, but Product Sound to this reference? can also be considered a part of the total experience - Is the Product Sound free from elements that un- using a product/system, which is a more marketing- dermine the total perceived quality? oriented approach. - Is the Product Sound acceptable in the environ- The overall objective in product development is to ment of the product? utilise future consumers’ attitudes, expectations, and More information can probably be obtained from preferences so that the sound from a product becomes a the marketing department which has often made – in positive attribute to the user instead of an annoying collaboration with external opinion research institutes problem. As all hearing persons can perceive acoustic – investigations of the users’ attitudes to and expecta- quality and thus can be said to be experts, there is a tions from certain product categories. Maybe there are great need for good acoustic design and development. even product developers and engineers in the firm whose knowledge can be used in connection with de- termination of the target sound. SOUND IN DESIGN A good product design combines functionality, ap- THE PRODUCT SOUND WHEEL pearance, and quality in an optimal way and adds to the pleasure of using the product. By consciously working We have created a model for optimising Product with the sound in the design an extra dimension of qua- Sound Quality, see Figure 1 [3]. lity can be added and even in some situations used di- The outer path in the Product Sound Wheel de- rectly as an important competitive parameter in mar- scribes the fundamental process of optimising the Pro- keting. The starting point of this work may be an ana- duct Sound Quality. First, alternative sounds from a lysis of the role of sound in the mental acts and product, simulated sounds, or sounds from similar pro- operations performed in interaction with the product. ducts are presented to a test panel. The panel gives its The object will be to examine the match between act response either in answering forms prepared for stati- and auditive feed-back. Is anything inappropriate and stical computations or directly, e.g. by setting sliders or to what degree ? [2] pressing buttons. The same sounds are measured by analysers, software, etc., and a number of metrics for In the future we expect that listening tests will be each sound is the result. The metrics may be any rele- used as a tool in product development to a higher de- vant traditional noise measure or may be more psycho- gree than now, both for objective auditive measure- acoustically related as loudness, sharpness, fluctuation, ments and for subjective measurements - affective tests strength, roughness, etc., or any combination of these. where preferences are asked for. By graphical or statistical methods the connections Furthermore we expect an increased use of sound in and correlations between the two kinds of measure- connection with simulation and virtual reality (VR); ments are sought, and usually it is possible to describe when designers by means of a VR Centre demonstrate the preferred sound by objective metrics. By analysis of new products, the sound is also evaluated in addition to the physical characteristics of the sound-generating me- the visual perception. The consequence of different chanisms, the necessary design changes to obtain the surfaces, joints, and materials can both be seen and defined values of the metrics may be implemented. heard! Tools for “sound tailoring or sound engineering”, sound To a greater extent marketing and market research editing, and simulation exist, and the lower inner path is on good Product Sound will take place via the Internet. often an attractive shortcut to test different versions of Sound Bars, where it is possible to hear the product possible sounds for further analysis or subjective tests. sound of various products before buying will pop up at A systematic approach in the design of low-noise certain retailers’. products is suggested in [4]. In a new 5-year project called “Human Sound Per- A properly designed product sound is an effective ception” DELTA will work intensively with research form of communication providing information about on some of these aspects and especially with how to quality, function, and condition of the product, espe- make measurements with test persons as an efficient cially as regards durable consumer goods. tool in the optimisation of perceived Product Sound. Product Sound can be advantageous to apply to: - enhance the perceived quality and even “brand REFERENCES sounding” – operational and signal sounds 1. J. Blauert & U. Jekosch, Sound Quality evaluation - support the user’s ego- - life style and – a multilayered problem, EEA-Tutorium, Ant- personal image werpen, 1996. - strengthen the position of the product in the mar- keting campaign 2. J. Bernsen, Sound in Design, Danish Design Center Further, manufacturers’ identity and image can be and DELTA, 1999. supported by the Product Sound – identity in a narrow 3. C. L. Fog, Optimal Product Sound: Design and sense and image in a wider sense. Construction Guidelines for Developing Products with Desirable Sound Characteristics and Minimal CONCLUSION Noise, Report SPM 144 (in Danish), DELTA, 1998. In summation, whether a product sound is attractive 4. ISO 11688: Acoustics – Recommended practice for is not determined by the sound alone and its relation to the design of low-noise machinery and equipment. function, but also by what the user is accustomed to, Part 1: Planning, Part 2: Introduction into physics what the competitors’ products do, and not least im- of low-noise design. portant: what the surroundings are willing to accept.

FIGURE 1. The Product Sound Wheel – a model for optimising Product Sound Quality.