Neutralizing the meaning of sound for sound quality evaluations H. Fastl Institute for Man-Machine-Communication, Technical University München, Arcisstr. 21, 80333 München, Germany Sound quality usually can be described by psychoacoustic magnitudes like loudness, 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 noise. 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 energy 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 pink noise (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 pure tone. 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 signal 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 sone 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 Acoustics” for support in realizing the sounds and editorial help. REFERENCES 1. Zwicker, E., Fastl, H., Psychoacoustics. 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 Hearing 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 signals: 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 frequency 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 noises 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].