psychoacoustic basis of sound quality evaluation€¦ · neutralizing the meaning of sound for...
TRANSCRIPT
Neutralizing the meaning of soundfor 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 anoriginal sound with pink noise. However, this procedure has the severe disadvantage that the original sound and the sound withno meaning differ in loudness. Therefore, a procedure is proposed, which can „neutralize“ the meaning of sound, and at the sametime preserve the original loudness.
INTRODUCTION
For the assessment of sound quality, psychoacousticmagnitudes like loudness, sharpness, fluctuationstrength or roughness have proven successful. In manycases, a combination of these basic psychoacousticmagnitudes can predict sound quality ratings bysubjects [1]. Despite the fact that the application ofthese principles usually leads to successful engineeringresults, sometimes the meaning of a sound mayconsiderably influence its evaluation.For example it could be verified both in the field (see[2]) and in the laboratory (see [3]) that at same energyequivalent A-weighted level, railway noise is preferredto road traffic noise. From an engineering approach itcould be shown [4] that spectral differences betweenrailway noise and road traffic noise are reflected in thespecific loudness patterns and can account partly forthe advantage of the railway noise. Also the time-structure of noise immissions from railway noise androad traffic noise differ significantly. In addition tothese differences, which can be assessed by physicalmeans, there may be also an influence of the differentmeanings of the sounds [3].
PROCEDURES
In order to neutralize the meaning of sound, in severallabs the following procedure is used (cf. Figure 1).From the original sound, the temporal envelope isextracted, e.g. by rectifying an low-pass filtering. In amodulator, this signal is multiplied with pink noise,leading to a sound which has the same temporalenvelope as the original sound but no meaning.However, because loudness depends on the bandwidthof a sound, even if the temporal envelope is faithfullyextracted and filled with pink noise, the original soundand the synthesized sound usually differ in loudness.
These differences are of great relevance, since theevaluation of sound quality in many cases cruciallydepends on the loudness of the sound ([1], [5]).
FIGURE 1. Traditional procedure to remove meaning ofsound.
The loudness differences to be expected because ofdifferent bandwidth are illustrated in figure 2. The leftpart of figure 2 shows the dependence of the loudnessof pink noise (PN) or an 1 kHz-tone on level. At samelevel, the loudness of the pink noise is much largerthan the loudness of the pure tone. For example at94 dB, as indicated by the vertical double arrow, theloudness of the pink noise is about a factor of 2.5larger than the loudness of the 1 kHz-tone.The right part of figure 2 shows original sounds (leftcolumn) and sounds with the same temporal envelope,filled with pink noise (right column). A comparison ofthe left and right column reveals that the temporalstructure of original and synthesized sounds is verysimilar. However, in many cases the loudness of thesynthesized sounds is larger. This discrepancy isillustrated in the left part of figure 2 by vertical lines.For example sound (a), when replaced by pink noisewith same temporal structure (b), increases fromaround 40 sone to around 60 sone. Likewise, the
simulations (d) and (f) of the sounds (c) and (e) show aloudness which is approximately 50% larger than theloudness of the original sounds.
FIGURE 2. Loudness differences of sounds with same levelbut different bandwidth.
Because loudness is such a dominant cue in soundquality evaluation, we looked for possibilities topreserve the temporal envelope and also the loudness.In this respect, the Fourier-Time-Transform (FTT [6],[7]) proved to be a very helpful tool. The principle ofthe procedure applied is outlined in figure 3. Theoriginal sound first is analysed by the FTT-procedure.After spectral broadening of the elements of the FTT-patterns, sounds are sythesized by an inverse FTTalgorithm. As a result we get a synthesized sound,which has the same temporal envelope, no meaningand – in contrast to the procedure illustrated in figure 1– also the same loudness-time function as the originalsound.
FIGURE 3. New procedure to remove meaning of sound.
Figure 4 shows for comparison original sounds andtheir counterparts with neutral meaning obtained by theprocedure illustrated in figure 3. When comparingloudness-time functions displayed in figure 4a vs.figure 4b, there are almost no differences discernible.This means that the proposed FTT-based procedure notonly faithfully reproduces the temporal envelope of the
original sounds, but also preserves their loudness ingreat detail.
FIGURE 4. Comparison of original sound (a) and corres-ponding sound without meaning (b).
OUTLOOK
With the tool illustrated in figure 3, we now have thepossibility to study sounds which are as much aspossible identical with the crucial difference that one isan original sound with a specific meaning, and theother is the corresponding synthesized sound, fromwhich the meaning was removed.
ACKNOWLEDGEMENTS
The author wishes to thank the members of his group“Technical Acoustics” for support in realizing thesounds and editorial help.
REFERENCES
1. Zwicker, E., Fastl, H., Psychoacoustics. Facts andModels. 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 therailway 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 vonOssietzky Universität Oldenburg, Oldenburg, 29-50,1996.
5. Fastl, H., Sound Quality of Electric Razors - Effects ofLoudness. In: Proc. inter-noise'2000, CD-ROM, 2000.
6. Terhardt, E., Fourier transformation of time signals: Con-ceptual revision. Acustica 57, 242-256, 1985.
7. Mummert, M., Sprachcodierung durch Konturierungeines gehörangepaßten Spektogramms und ihre Anwen-dung zur Datenreduktion. VDI Reihe 10, Nr. 522, VDIVerlag, 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 physicalmetrics which show good correlation with subjective impression in order to predict the sound quality and find appropriatecountermeasures. In this paper, the validity of physical metrics of sound quality will be discussed on the basis of the results ofpsychological experiments, especially focusing attention on the temporal aspects of sounds.
INTRODUCTION
Much effort has been made to reduce the sound levelof machinery noise. However, since there is a limit toreduce the sound level and machinery noises give usinformation concerning the situation of machines,recently it is expected to improve sound quality ofmachinery noise as well as to reduce the sound level.It is important to find physical metrics which showgood correlation with subjective impression in order topredict the sound quality and find countermeasures toimprove sound quality. Many physical metrics havebeen proposed [1]. Some of them usually show goodcorrelation with subjective impression, but others donot always. Validity of physical metrics of soundquality will be discussed on the basis of the results ofpsychological experiments, especially focusingattention on the temporal aspects of sounds.
DIMENSIONS OF SOUND QUALITY
Sound quality is multi-dimensional. However in mostof our former studies [2], three main factors, powerful,metallic and pleasant, have consistently been extractedand they can be regarded as representative factors ofsound quality. An example of the result of factoranalysis is shown in Table 1 [3].
RELATION BETWEEN PHYSICALMETRICS AND DIMENSIONS OF
SOUND QUALITY
Coefficients of correlation between physical metricsand adjective scale values are shown in the rightcolumns in Table 1. Good correlation is usually foundbetween LLz (Zwicker’s loudness level based on ISO532B averaging temporal fluctuation on energy basis)and the impression of powerful factor [4]. LAeq alsoshows good correlation with the impression ofpowerful factor when sounds have broad band
frequency components. When the sound contains highfrequency components, it causes the impression‘sharp’ and ‘metallic’. In this case, the impression ofmetallic factor can be evaluated by calculatedsharpness [1, 5]. Pleasant factor is related to cognitiveand cultural factors as well as physical properties ofsounds. It is difficult to predict pleasantness of soundsby physical properties alone. However, in a limitedsituation, it may be possible to find physical propertieswhich shows good correlation with subjectiveimpression. For example, equal pleasantness contourfor air-conditioner noise was proposed based on LLzand calculated sharpness [6].Hearing is a sensation which conveys informationalong temporal stream. Therefore, temporal factorshave an important effect on hearing. The temporalpattern and temporal condition of sounds were foundto have a significant effect on each dimension of soundquality in our former studies.
(1) Temporal condition and powerfulfactor
The loudness is usually evaluated by LAeq or LLz as thefirst approximation. However, when the temporalpattern of the sound is systematically varied, the soundwhich has high level portion at the beginning isperceived as being louder [7]. This may be due to theovershoot at the onset of the sound.
(2) Temporal condition and sharp factor
The impression varies according to the duration ofsounds [8]. It was found that the shorter the durationbecame, the more sharp the sound was judged asshown in Fig.1.
(3) Temporal condition and pleasant factor
An example of the effect of temporal pattern on
pleasantness is shown in Fig.2. Temporal changes ofcalculated sharpness of the sounds when a golf ballwas hit by a golf club were found to be different. Therelation between subjective impression of refresh andthe temporal change of sharpness is shown in Fig.2.High correlation between them suggests that thetemporal change has a significant effect on subjectiveimpression. When pulse train is judged, the intervalbetween pulses has a significant effect on theimpression. A phrase of a music performance(Pictures at an Exhibition composed by Musorgsky)was played with synthetically varying the intervalbetween sounds. It was found that the impressionchanges systematically with the interval betweensounds [9]. Similar results were found with gear noisein a car [10].
FINAL REMARKS
There are various factors which contribute to the soundquality. Temporal factor is one of them and it has a
significant effect on the impression of sounds even ifother physical properties are equal. The temporaleffect may be related to the dynamic characteristics ofhearing. It is important to find appropriate method toevaluate sound quality taking temporal factors as wellas other physical properties of sounds intoconsideration.
REFERENCES
1. E. Zwicker and H. Fastl, Psychoacoustics,, (Springer, 1999).2. S.Namba, Measurement of Timbre and its Applications,
(Oyogijutu Shuppan, 1992).3. S. Namba, et al., J.A.S.J. (E), 13, 49-58 (1992).4. S. Kuwano, et al., Noise Cont. Eng. J., 33, 107-115 (1989).5. G. von Bismarck, Acustica, 30, 159-172 (1984).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).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
University, 41, 17-35 (1993).10. T. Abe, Dissertation, Osaka University (1995).
Table 1 Result of factor analysis
Adjectives Factor 1 Factor 2 Factor 3 r (LLz) r (sharpness)gentle – soft
distinct – dullnoisy– quiet
deep – metallicpure – impure
loud – softcalm – shrill
harmonic – discordantpleasant – unpleasant
powerful – weakflat – rumblingsmooth - harsh
.725-.615-.342.878-.371-.139.842.553.374-.073.016.788
-.114.425-.387-.036.660-.148.044.445.528-.054.611.186
-.224.229.546-.051-.040-684-.160-.060-.204.667-.223-.161
.118
.120-.707.041.362-.819.086.321.712-.834.549.291
.796-.722-.742.831-.571-.590.871.838.737-.582-.200.892
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
Various methods are at hand for sound quality as-sessment, and at the present time the acoustic engineer is in the position to employ modern signal analysis tools for the calculation of elementary psychoaoustic magnitudes such as loudness, sharpness, or roughness. However, since the perception of sounds is dependent on cognitive and emotional factors as well, additional measurements are needed to get the whole picture of sound quality.
For decades the use of the semantic differential technique, which Osgood [1] developed to identifiy emotional meanings of words, has been extended to a large variety of different concepts, including sounds [2]. A semantic differential (SD) comprises several pairs of opposite adjectives which constitute the poles of mostly 7-point bipolar rating scales (see figure 1). An important methodological issue is that a disctinc-tion has to be made between connotative and denota-tive scales. In the present study, the latter refers to acoustic or psychoacoustic properties of the sounds such as loudness. By contrast, connotative scales are intended to measure the emotional meaning contained in the sound on scales such as “calming – exciting”.
In the present study, an SD comprising both types of scales is applied to a series of every-day noises in order to treat the following issues: • Test-retest reliability of the method • Factors of sound quality (underlying the SD) • Correlation of denotative scales with psychoacoustic
calculations (signal analysis)
EXPERIMENTS
Method
Subjects
A total of 21 subjects (10 male, 11 female) participated in the first experiment. The sample mainly consisted of students 20 to 56 years of age (median: 26). An inde-pendent second sample of 21 students (5 male, 16 female), ranging from 19 to 31 years (median: 25), was recruited for a retest six month later. All subjects re-ported normal hearing.
Stimuli
Stimuli were 17 environmental sounds which stem-med from a broad range of sources such as musical instruments, natural environment, household appli-ances, and power tools. The series contained both sta-tionary and level fluctuating noises (duration: about 5 seconds). Levels ranged from 61 to 84 db(A) Leq, which were intended to reflect natural loudness ratios.
Procedure
The task required the subjects to judge the sounds,
which were randomly presented through headphones, with a set of 20 bipolar rating scales as depicted in figure 1. The presentation of each sound was looped until all scales had been completed. Prior to the judge-ments, subjects were presented with all sounds for orientation.
Results and Discussion
Reliability
For both groups of subjects (test and retest), individual judgements were averaged over the adjective scales. Thus, for each sound two mean profiles result, as ex-emplified in figure 1. Test-retest reliability of the 17 sounds ranges from r=.79 to r=.97, and amounts to r=.95 for the whole stimulus set. Since no substantial differences between the two groups resulted, both datasets were aggregated for further analysis.
3 2 1 0 1 2 3
unpleasant
flat
muffled
dark
peaceful
ugly
sad
low
soft
light
calming
smooth
pure
gentle
dull
slow
weak
boring
unsteady
soft
3 2 1 0 1 2 3
pleasant
rumbling
shrill
light
aggressive
beautiful
bright
high
loud
heavy
agitating
rough
impure
harsh
sharp
fast
strong
exciting
steady
hard
3 2 1 0 1 2 3
unpleasant
flat
muffled
dark
peaceful
ugly
sad
low
soft
light
calming
smooth
pure
gentle
dull
slow
weak
boring
unsteady
soft
3 2 1 0 1 2 3
pleasant
rumbling
shrill
light
aggressive
beautiful
bright
high
loud
heavy
agitating
rough
impure
harsh
sharp
fast
strong
exciting
steady
hard
3 2 1 0 1 2 3
unpleasant
flat
muffled
dark
peaceful
ugly
sad
low
soft
light
calming
smooth
pure
gentle
dull
slow
weak
boring
unsteady
soft
3 2 1 0 1 2 3
pleasant
rumbling
shrill
light
aggressive
beautiful
bright
high
loud
heavy
agitating
rough
impure
harsh
sharp
fast
strong
exciting
steady
hard
FIGURE 1: Mean profiles (test� and retest�) for stimulus “hair dryer” (retest-reliability: r=.92). Standard deviations (all noises) ranged from 0.15 to 2.21 scale units.
Factor Analysis
Varimax rotated principal component analysis was
employed to extract orthogonal factors underlying the 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 to the factors as follows: • Factor 1 “Evaluation” (29 %): ugly-beautiful, un-
pleasant-pleasant, calming-agitating, boring-exciting, gentle-harsh, pure-impure, soft-hard
• Factor 2 “Timbre” (17 %): dark-light, low-high, muffled-shrill, dull-sharp, light-heavy
• Factor 3 “Power” (16 %): weak-strong, soft-loud, flat-rumbling
• Factor 4 “Temporal change” (8 %): unsteady-steady, smooth-rough
Correlations with Psychoacoustic Calculations
For each sound, average loudness (Zwicker), sharp-ness (Aures), and roughness were calculated by use of commercially available signal analysis software (Head acoustics Artemis). The following coefficents result for the correlations with the respective adjective scales: • roughness vs. “smooth-rough”: r=.84 (p<.01) • sharpness vs. “dull-sharp”: r=.71 (p<.01) • loudness vs. “soft-loud”: r=.71 (p<.01)
In addition, loudness calculations were correlated with judgements of a previous study [3], in which the same sounds had been judged using category subdivi-sion (CS) scale. The CS-scale comprises five verbally distinguished categories (“very soft”, “soft”, “me-dium”, “loud”, “very loud”), with a 10-step fine graduation in each category. The correlation between the mean values on this so-called 50-points scale and loudness (Zwicker) was r=.91. This correlation dif-fered significantly from that of the 7-point scale (p<.05) with loudness.
CONCLUSIONS
Test-retest reliability (r=.95) of mean semantic pro-files indicates high accuracy of the semantic differen-tial technique. The factorial investigation of the 20 adjective scales revealed four orthogonal factors which were interpreted in terms of “evaluation”, “tim-bre”, “power”, and “temporal change”.
Denotative scales of the SD showed high correla-tions with respective psychoacoustic calculations (loudness, sharpness, roughness), as expected. Com-pared to the 7-point adjective scale, a significantly higher correlation for loudness was found with the 50-points category subdivision scale.
ACKNOWLEDGEMENTS
The authors are grateful to Andrea Hahn, Petra Schüller, and Mirjam Wolf for their contributions in this study.
REFERENCES
[1] Osgood, C. E., Suci, G. J., and Tannenbaum, P. H. The
measurement of meaning. University Press of Illinois. Urbana 1957.
[2] Schick, A. Zeitschrift für Lärmbekämpfung 41(3), 61-68 (1994).
[3] Zeitler, A. and Hellbrück, J. Psychophysical scaling of the pleasantness of environmental sounds, in Proceedings of the 7th Intl. Congress on Sound and Vibration. Gar-misch-Partenkirchen, 2000, pp. 2485-2490.
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 background noise do not form clearly differentsound sensations any more, and we observed that different subjects focused their attention towards different tonal components whilerating the Sound Quality. The results derived from experiments with single tonal components can thus not be transferred to the caseof multiple tonal components, and special attention has to be paid to the method to evaluate Sound Quality of these signals.
INTERIOR VEHICLE SOUND
The evaluation of the Sound Quality of interior ve-hicle sounds is a complex task. First, the sound is com-posed of a variety of different components, and second,the perception of Sound Quality by humans is a com-plex process. This process is not only based on the purephysical signal, but also on other modalities like visualor tactile information and even non-sensory moderators(e.g., [1]).
In general, a product sound consist of two differentgroups of features, undesired and desired features. Thefirst group comprises the sound features which have tobe avoided (e.g., squeak and rattle), the second groupcomprises the sound features which form a „good“sound (e.g., pleasant, sporty, noble).
The undesired features play an important role, be-cause they can significantly degrade a good sound andspoil the effort spent to create it. They have to be redu-ced to that level where they do not degrade the overallsound (acceptance level).
But, this task is complicated by the fact that the clas-sification of specific components might change. Todayfor example the characteristic sound of a turbo chargeris not classified as undesired, because it gives the feed-back that the car is equipped with this non-standard fea-ture. But in the future, when nearly all vehicles with adiesel engine will be equipped with a turbo charger, thisfeedback information will loose its importance, and thesound might be classified as undesired.
Interior vehicle sound is composed of the majorcomponents from engine, road, and wind, plus contri-butions from various components. Since the overall le-vel and undesired sounds like squeak and rattle havebeen significantly reduced in the past, now other soundcomponents turn out to be undesired. One important
class are tonal components, which are caused by va-rious vehicle components:
• the engine sound consists of multiples of the engi-ne order
• road noise might include tonal components fromtires or road surfaces
• the transmission often creates tonal components• automatic steering, oil and fuel pumps, chains, and
all kind of electric motors cause further contributi-ons.
Fig. 1 shows that an interior vehicle sound is compo-sed of multiple tonal components covering a broad fre-quency range. It is important to note that thecomponents often do not show ideal sinusoidal charac-ter, but can be much broader.
TONAL COMPONENTS
It is obvious that the evaluation of the influence ofthe various tonal components is a difficult task, althou-gh the investigations on the perception of tonal com-ponents have a long tradition. Basic psychoacousticresearch data is available from literature with respect todetection and masking (e.g., [2]), but they basically allconsider either the case of a single tonal component in abroadband, non-tonal masker, or the relation of two to-nal components without any masker. The descriptionabove shows that these data can thus not directly be ap-plied to interior vehicle sounds.
Furthermore, in the context of Sound Quality not thedetection of tonal components is important, but their ac-ceptance level. We thus planned several experiments toinvestigate the frequency-dependency, the role of thewidth of a component, and the influence of multiple to-nal components.
All experiments have been conducted in a sound-
proofed chamber using headphone representation andapplying the individual test [3]. They have been basedon real interior vehicle recordings of several vehicleswhere the existing tonal components have been modi-fied by filtering or synthetic sounds have been added.Psychoacoustical investigations have been conductedfor the following parameter variations:
• frequency variation of a single tonal component• combination of two tonal components • combination of three tonal components• a synthetic ideal sinusoidal tonal component• a real broad tonal component (bandwidth 80 Hz
around 800 Hz).A group of experts and a group of non-experts parti-
cipated in the tests and determined the detection and ac-ceptance thresholds and rated the strength of the whine.The findings can only be summarized here:
• the expected frequency-dependency is reproduced:higher frequency components are more annoyingthan lower frequency components (e.g., [2])
• the difference between detection and acceptancethreshold depends on frequency and is about 6 dBat low frequencies and 3 dB at high frequencies
• experts render reproducible and stable results• non-experts have problems to rate whine in a re-
producible manner. The rating of a subset of sti-muli rated two times depended on the other stimulipresented in the respective test (context effect)
• broader components also causes tonal sensationscomparable to sinusoidal components, but thespectral peak level is not the appropriate descriptor
• the presence of multiple tonal components influ-ences the whine rating. It seems that contributionsof several components are combined to produce awhine rating.
A very important and critical finding turned out in anexperiment where gear whine was investigated. There
Fig. 1 Typical narrowband spectrum of an interior vehiclesound (idle at 1200 rpm, red=right, green=left chan.)
the sounds of different vehicles once were rated whenonly the condition with whine was presented, and inanother test with a direct comparison of the situationswith and without whine (clutch engaged and disenga-ged). Differences in the ratings could be observed: itturned out that in the first case the overall whine was ra-ted, while in the latter case the attention of the listenerwas automatically focused to the gear whine, so thatonly that whine component was rated.
But, even the rating of the overall whine impressionwas not stringent for all subjects. In a discussion it tur-ned out that a kind of global tonal impression arises, butthat in addition usually one prominent tone was percei-ved caused by a single tonal component. But, differentsubjects detected different tones when listening to thesame stimuli. Furthermore, if this prominent tonal com-ponent was eliminated, in a first instance the whine im-pression was reduced, but after listening to it for sometime, another tone seemed to „pop out“ of the back-ground noise and could be heard as a single component.
The perception of tonal components in these com-plex stimuli thus does not only depend on pure physicalparameters - it also depends on cognitive aspects, herein the form of attention. The attention of a subject is fo-cused on one of the components present in the stimuli.
This finding showed that most of the results of basicperceptual investigations of tonal components can notdirectly be applied to interior vehicle sound. In most ba-sic experiments the level of one tonal component wassystematically varied, so that the attention of the sub-jects was automatically focused to that component andthe contribution of other components was suppressedby the listener. In a complex sound like the interior ve-hicle noise the perceptual process is different, since theattention of the subjects is not automatically focused.
SUMMARY
The approach to adopt basic psychoacoustic data de-rived from experiments with single tonal componentsby superposition of the effects can not be applied to theperception of multiple tonal components. In this com-plex condition non-acoustic moderators influence theperception of humans.
LITERATURE
1. Bodden, M., “Perceptual Sound Quality Evaluation“, inProc. Internoise 2000, Nice, France, 2000.
2. Zwicker, E., and Fastl, H., Psychoacoustics -Facts andModels, Springer Verlag, 1990.
3. Bodden, M., and Heinrichs, R., “Evaluation of interior ve-hicle noise using an efficient psychoacoustic method“, inProc. of the Euronoise 98, 1998.
The influence of sound on perception thresholds and JNDs of whole-body vibrations
R. Weber, I. Baumann, M. Bellmann and V. Mellert
Department of Physics - 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
Whole-body vibrations usually have negative effects on subjective comfort perception which becomes increasingly important in transportation vehicles. Only few basic research is published on the perception of whole-body vibrations in vehicles. Related norms as e.g. ISO 2631 1/2 [3, 4], refer to whole-body vibrations in buildings. There subjects are seated on rigid chairs that differently transfer vibrations compared to car seats. This study investigates perception thresholds and just noticeable level differences (JNDs) of vertical whole-body vibrations on a real car seat. Additional influence of (acoustical) noise on threshold is regarded.
EXPERIMENTAL SETUP
Using an adaptive 3-AFC 1up-2down-procedure perception thresholds and JNDs of vibration levels (70.7 % point of the psychometric function) are determined on a new car seat. Sinusoidal test frequencies vary from 12.5 Hz to 80 Hz in third octave intervals. Signals of 1000 ms are seperated by pauses of 500 ms. Each threshold measurement is repeated three times by each subject on three different days. Signal intervals are optically marked on a computer monitor. The car seat is excited vertically by using a „sound and vibration reproduction system “, developed by the ITAP GmbH in cooperation with the ACOUSTICS group [1, 5]. It is mounted on a decoupled basement in a sound proof room with a acoustic background level of 38 dB(A). The device is optimised with respect to diminish sound radiation during vibration.
PERCEPTION THRESHOLDS
12 subjects (6 male, 6 female / 22-32 years old) participate in the adaptive tests (starting step size 8 dB / minimal step size 1 dB). The dashed curve in figure 1 shows means of perception thresholds as a function of frequency in the case of no additional acoustic signal. Results are given in terms of acceleration levels LVib in [dB] and acceleration amplitudes a [m/s2] (LVib. = 140 dB corresponds to a = 10 m/s2).
12,5 16 20 25 31,5 40 50 63 8070
75
80
85
90
95
100
10 100
frequency [Hz]
acc
leve
l [dB
]
with pink noisewithout pink noise
0,0056
0,01
0,0178
0,0316
0,0562
0,1
0,0032
acc.
am
pl. [
m/s
²]
FIGURE 1. Means and standard deviations of perception thresholds of vertically excited whole-body vibrations on a real car seat (8 subjects) with (-- -- --)
and without (-----) masking (acoustic) pink noise of 68 dB(A).
Starting with a level of 82 dB the perception threshold increases with a slope of about 5 dB/octave up to 50 Hz whereas an unexpected decline is observed up to 80 Hz. To check whether sound radiated from the vibration pad has triggered the subject’s responses and hence is 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 (solid line in figure 1) turn out to be on average higher by 1.4 dB. However, they also exhibit this unexpected decline for ‘higher’ vibration frequencies. Hence, this decline cannot attributed to noise emitted by the vibrating pad.
When comparing perception thresholds of Miwa 1969, McKay 1971, Parsons & Griffin 1988 (extracted from [2]) and perception thresholds by Bellmann et al. [1] with our own data essentially similar frequency dependencies can be noted (see figure 2). Considerable discrepancies exist with current norms [3, 4].
12,5 16 20 25 31,5 40 50 63 8070
75
80
85
90
95
100
10 100frequency [Hz]
acc.
leve
l [dB
]
Miwa, 1969McKay, 1971Parsons&Griffin, 1988Parsons&Griffin, 1988Bellmann et al., 2000ISO 2631 (z)own data, without noiseown data, with noise
0,01
0,0316
0,1
FIGURE 2. Comparison of literature data [1,2] on perception thresholds of whole-body vibrations and norm data [4] with own thresholds with and without masking noise.
JUST NOTICEABLE DIFFERENCES
10 subjects ( 6 male / 2 female) take part in determining just noticeable differences of vibration level. Starting level in the adaptive 3 AFC trials is 110 dB (= 0,316 m/s²), reference signal level is 100 dB (= 0,1 m/s²), starting step size is 4 dB and minimal step size is 0.5 dB. Resulting JND-Ls (figure 3) show frequency dependent means of about 1.6 dB with standard deviations von 0.46 dB. Nearly no differences can be observed. The results fit well to former test results obtained on a hard chair [1].
12,5 16 20 25 31,5 40 50 63 800,0
0,5
1,0
1,5
2,0
2,5
3,0
10 100frequency [Hz]
leve
l diff
eren
ce [d
B]
FIGURE 3. Means (thick solid) (std-devs included) and medians (dashed) of just noticeable differences of vertically excited whole-body vibrations on a real car seat (8 subjects) with a masking (acoustic) pink noise of 68 dB(A). Curve (thin solid) shows (rigid chair-) data from [1].
CONCLUSIONS
Perception thresholds of vertically excited sinusoidal whole-body vibrations (background noise level 38 dB) 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 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 the ‘silent’ case. Just noticable differences of whole-body vibration level JND-Ls are 1.6 dB independent of frequency from 12.5 Hz to 80 Hz. They are consistent with available literature results [1].
REFERENCES
1. Bellmann M.A., Mellert V., Reckhardt C. und Remmers H., “Experimente zur Wahrnehmung von Vibrationen” In: Fortschritte der Akustik, DAGA 2000, 2000
2. Griffin M. J., “Handbook of human vibration”, Academic Press, 1991
3. ISO 2631-1, “Evaluation of human exposure to whole-body vibration – Part 1: General requirements”, International Organization for Standardisation, Geneva (1997)
4. ISO 2631-2, “Evaluation of human exposure to whole-body vibration – Part 2: Continuous and shock-induced vibration in buildings (1-80 Hz)”, International Organization for Standardisation, Geneva (1989)
5. Remmers H. & Bellmann M.A., “System zur realistischen Wiedergabe von Schall und Vibrationen”, In: Fortschritte der Akustik, DAGA 2000, 2000
Psychoacoustic sensation magnitudes and sound qualityratings 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 differentbrands, 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 thosesounds will be discussed. It will be challenged which psychoacoustic sensations are instrumental for the preference of the soundquality of a specific car.
INTRODUCTION
Standing next to a car, the idling noise of it is in mostcases a sufficient hint on the kind of motor. In contrastto a gasoline powered car, a diesel powered car showsa typical sound character which will be called in thefollowing "dieselness".To reveal the correlation between this characteristicsound described by dieselness and the basic psycho-acoustic sensation magnitudes loudness, sharpness,roughness and fluctuation strength were investigated inpsychoacoustic experiments for eight different outdooridling noises. Among those eight cars were a gasolinepowered car ("gp"), four diesel powered cars ofdifferent brands ("dp b1" to "dp b4")as well as threedifferent adjustments of the motor of one particulardiesel powered car ("dp a1" to "dp a3").Furthermore, the hypothesis was posted that thisspecial sound character of diesel powered cars(dieselness) is responsible for the judgement on thesound quality of the car. Therefore, by means of aranking experiment the sound quality of those eightoutdoor idling noises was assessed additionally.
EXPERIMENTS
The outdoor idling noises of the above mentioned carswere recorded by a dummy head system of HEADAcoustics positioned at a distance of 1 m lateral to theright front wheel at a height of 1.70 m. For theexperiments, the sounds were presented in a sound-proof booth via a freefield equalized [1] STAX head-phone calibrated to reproduce the original sound level.To evaluate the loudness, sharpness, roughness,fluctuation strength and dieselness of the sounds, themethod of "magnitude estimation with anchor sound"was used, and the sound of the diesel powered car of
brand 1 ("dp b1") acted as anchor sound. The soundquality was assessed with a ranking method which isdescribed in detail in [2]. Fourteen normalhearingsubjects with a median age of 27.5 years (4 female, 10male) participated in the experiments.
RESULTS
Figure 1 shows the medians and the interquartileranges of the sensation magnitudes loudness (squares),sharpness (triangles), roughness (circles), fluctuationstrength (rhombs) and dieselness (stars) for thegasoline powered car, the four different brands ofdiesel powered cars and the three different motoradjustments of the diesel powered car.
0
50
100
150
200
250
gp dp b1
dp b2
dp b3
dp b4
dp a1
dp a2
dp a3
rela
tive
se
nsa
tio
n m
ag
nit
ud
e /
%
loudness sharpness
roughness fluctuation strengthdieselness
FIGURE 1. Results for the sensation magnitudes loudness,roughness, sharpness, fluctuation strength and dieselness.
If the relative sensation magnitudes of the gasolinepowered car are compared with those of the dieselpowered cars of different brands, it can be stated, thatthe gasoline powered car produces about 65 % of theloudness, 50 % of the sharpness, 60 % of theroughness, 30 % of the fluctuation strength and only
10 % of the dieselness of that diesel powered car withthe respectively lowest estimated sensation magnitude.Comparing the results of the four different brands,roughness is the sensation magnitude which is varyingmost (about 50 percentage points) and fluctuationstrength less (about 15 percentage points). Thedieselness is differing between the four differentbrands for maximum 32 percentage points.With the investigated three adjustments of the motor,fluctuation strength can be changed less (about 35percentage points) and dieselness most (about 120percentage points). In loudness, sharpness androughness a difference of 85 to 100 percentage points,in fluctuation strength of about only 35 percentagepoints can be obtained.Table 1 shows the rank correlation coefficients(according to Spearman) between the sensationdieselness and all other magnitudes. The correlation isfor all magnitudes very high but best betweendieselness and roughness (ρ = 0,976).
TABLE 1. Rank correlation coefficients ρ betweendieselness and the other sensation magnitudes loudness (N),sharpness (S), roughness (R) and fluctuation strength (F).
N S R F0,970 0,952 0,976 0,857
Figure 2 shows the median and the interquartile rangesof the ranks given in sound quality for the eight carsinvestigated. In judging the car with the best (gasolinepowered car) and the worst (diesel powered car withmotor adjustment 3) sound quality subjects judgedconsistently. Whereas the diesel powered car withmotor adjustment 2 is rated better than all other dieselpowered cars, that one with motor adjustment 1 isclassified behind brand 1 and 2 but still before brand 3and 4.
12345678
gp dpb1
dpb2
dpb3
dpb4
dpa1
dpa2
dpa3
ran
k
FIGURE 2. Results for the sound quality ratings.
DISCUSSION
Figure 3 shows the results sorted with descendingsound quality: an increase in all sensation magnitudesseems to go in line with the deterioration in soundquality. This can also be confirmed by the rank
correlation coefficients (table 2) between the rank insound quality and the sensation magnitudes: in allcases a strong correlation is given but especially thesensation dieselness (ρ = 1) seems to be an importantclue for the subjects in classifying the sound quality.
0
50
100
150
200
250
1 gp
2 dp a2
3 dp b2
4 dp b1
5 dp a1
6 dp b4
7 dp b3
8 dp a3
rela
tive
se
nsa
tio
n m
ag
nit
ud
e /
%
loudness sharpness
roughness fluctuation strength
dieselness
rank
FIGURE 3. Results for the sensation magnitudes orderedwith respect to their sound quality.
TABLE 2. Rank correlation coefficients ρ between the rankin sound quality and the sensation magnitudes loudness (N),sharpness (S), roughness (R), fluctuation strength (F) anddieselness (D).
N S R F D
0,970 0,952 0,976 0,857 1,000
CONCLUSION
The outdoor idling noise of the gasoline powered car,the diesel powered cars of different brands andespecially the different adjustments of the motor of oneparticular diesel powered car differ substantially interms of the psychoacoustic sensation magnitudesloudness, sharpness, roughness and fluctuationstrength.The sensation characterizing the typical sound of adiesel powered car - here called "dieselness" - is highlycorrelated with those basic psychoacoustic sensationmagnitudes. Furthermore, the strength of the sensationdieselness seems to be the cause how subjects rank thesound quality of the sound.
REFERENCES
1. Zwicker, E., Fastl, H., Psychoacoustics - Facts andModels. 2nd updated ed., Springer Verlag, Berlin 1999.
2. Patsouras, Ch., Fastl, H., Patsouras, D., Pfaffelhuber, K.,Subjective evaluation of loudness reduction and soundquality ratings obtained with simulations of acousticmaterials for noise control, in Proceedings of Euronoise2001, edited by Demos Tsahalis, CD-Rom, 2001.
Effects of Modulation on the Quality of Diesel Engine NoiseA. 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 amplitude 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 combustiontiming and amplitude in each of the engine’s cylinders. Measurements of cylinder pressure during combustion are also incorporated.Deterministic and random variations in the timing and amplitude of the combustion processes were simulated at different levels tocreate a set of sounds with varying degrees of modulation. Subjective and objective (roughness and fluctuation strength) evaluationswere used to examine the relationship between overall diesel engine sound quality and the degree of variability in timing and amplitude.
INTRODUCTION
Diesel engines are simpler, have better fuel economy,and are more durable than gasoline engines. However,people complain about diesel engine noise and oftenrate it as being more objectionable than gasoline enginenoise. Diesel engine noise may contain strong modula-tions, tonal components and sound impulsive [1, 2, 3].Here the focus is on the effects of modulations. A modelhas been developed to synthesize diesel engine soundsso that the relationship between modulations, caused bytiming and amplitude variations, and the sound quality ofdiesel engine noise can be examined.
SYNTHESIS
The main focus of the research described in this paperis to gain a better understanding of how variation in com-bustion timing and combustion pressure amplitude affectthe perceived sound quality of diesel engines. Becauseit is difficult to separate the individual combustion eventsin a measured noise signature, these effects are being in-vestigated by synthesizing the sound of the combustionevents starting with an impulse train for each cylinder inthe engine. The main components of the synthesis are:
• Generation of a no-variation (NV) uniform impulsetrain for each cylinder based on engine revolutionsper minute (rpm).
• Vary the timing and amplitude of each impulse train.
• Convolve the impulse trains with combustion pres-sure profiles.
• Filter the synthesized combustion pressures to simu-late the relationship between combustion and acous-tic pressure.
Impulse train
The first step in synthesizing the sounds is to gener-ate an impulse train for each cylinder that is based onthe engine rpm and the total number of cylinders (Nc).The impulse train template for four stroke engines has aperiodicity dependent on the engine speed and the num-ber of cylinders, as shown in Equations 1 and 2, whereTSingleCylinder is the timing between consecutive impulsesin a cylinder and Teng is the time between impulses in theengine.
TSingleCylinder� 120
rpmseconds � (1)
and
Teng� TSingleCylinder
Nc� (2)
The template has its amplitude and timing varied foreach cylinder. This variation has both a random anda fixed component. The fixed component causes eachcylinder to be slightly different from the others. The ran-dom component causes individual combustion events tobe slightly different from the others, as might be the caseif the combustion event is not truly repeatable and timingcannot be tightly controlled.
The fixed timing variation, starts by providing an off-set so that the nth cylinder fires
�n � 1 ��� Teng seconds after
the first. This time is then modified by a percent of Teng.The variation is again altered, this time randomly. Thisrandom variation has a uniform distribution with a maxi-mum level specified to be a particular percentage of Teng.The amplitude variation is done in a similar fashion, ex-cept that there is no initial offset and the modifications arebased on a percent of the template amplitude, not Teng.
0 0.02 0.04 0.06 0.08 0.10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time, seconds
Pre
ssur
e/m
ax(P
ress
ure)
, bar
/bar
FIGURE 1. The combustion impulse response. The pressurehas been normalized to a peak amplitude of one.
Combustion and system response
To convert an impulse train into a set of combus-tion events, a combustion impulse response is used. Asingle cylinder diesel engine’s combustion pressure wasrecorded and averaged over 50 cycles. The result for the 1cylinder engine operating at 1200 rpm is shown in Figure1. This averaged pressure profile was used as the basis forthe combustion impulse response. Because this pressureprofile changes with engine speed, the synthesis routinecompresses or expands this profile in time for higher orlower, respectively, engine speeds.
To convert the set of combustion events into actualsound, a filter is needed to simulate the paths from thecylinders to the listener. In the current study a simplelowpass filter is used and is the same for all cylinders.Future refinements of the simulation will include morerealistic models of these paths.
SOUND QUALITY ANALYSIS
In order to determine how timing and amplitude varia-tion effect sound quality, sounds were generated with dif-ferent levels of fixed and random variation applied to theamplitudes and timings of the impulse trains. These vari-ations ranged from zero to fifty percent of the amplitudeand the Teng values. The rpm of a six cylinder engine wasset at 700, making the nominal no-variation timing be-tween engine combustion events 0.0286 seconds. Afterall sounds were adjusted to the same loudness level, met-rics were calculated and the sounds were also evaluatedsubjectively.
It was found that the roughness and fluctuationstrength metrics were sensitive to the type as well asthe level of the variation. Example results for 10% and50% variations are shown in Figure 2. Timing varia-
NV FA RA FT RT2.3
2.4
2.5
2.6
2.7
2.8
2.9
3
3.1
3.2
3.3
Rou
ghne
ss L
evel
, asp
er
10% Variation50% Variation
NV FA RA FT RT0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
Flu
ctua
tion
Str
engt
h, v
acil
10% Variation50% Variation
FIGURE 2. A comparison of modulation metrics with 10% and50% variation. NV denotes No Variation. F and R denote thefixed and the random method of variation. A and T denote vari-ation applied to amplitude and timing.
tions tended to decrease roughness and increase fluctu-ation strength. To a lesser extent, amplitude variationsincreased fluctuation strength, but did not have a strongimpact on roughness.
The results of the metric calculations were confirmedby the subjective analysis. When listening to the sounds,it was found that sounds with the fixed variation appliedto either amplitude or timing had a stable modulated qual-ity. When timing was varied, the modulations were per-ceived to be much stronger than modulations resultingfrom amplitude variations at the same level. Sounds withthe random variations also had a modulated quality butsounded more erratic. Mirroring the objective analysis,sounds with timing variations sounded less rough than theoriginal sound with no variation.
ACKNOWLEDGMENTS
The Authors would like to thank Isuzu Motors for theirsponsoring of this research and Lijun Song and Dr. JohnAbraham of the Zucrow Laboratories for their assistancein acquiring initial combustion profiles.
REFERENCES
1. R. Ingham, N. Otto and T. McCollum, The Society of Auto-motive Engineers. 1999-01-1819, 1295-1299 (1999).
2. H. Takata, T. Nishi and P. Davies, Proceedings of Inter-Noise 99, Fort Lauderdale, Florida, 1201-1206 (1999)
3. M. Russell, S. Worley and C. Young, The Society of Auto-motive Engineers. 870958, 79-95 (1987).
Effect of Factors other than Sound to the Perception of Sound Quality
T.Hashimotoa and S. Hatanoa
aDepartment of Mechanical Engineering, 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 affected by various parameters such as visual scenery, image of a car, seat/floor vibrations etc[1,2]. In order to see these effects quantitatively, two types of subjective evaluation test were conducted using SD method. The one is to see the effect of image a car and the other is to see the effect of visual scenery and seat/floor vibrations to the perception of sound quality.
SD EXRERIMENT TO SEE THE EFFECT OF IMAGE OF A CAR
The image of a car affects significantly the result of subjective evaluation of sound quality of car interior noise. In order to see this effect, SD experiments were conducted.
The effect of an image of a car was tested by SD method. Subjects were 20 males and 20 females aged between 18 and 25 years. As was shown in Fig.1, the evaluation only by moving image was the best and that obtained only by noise exposure was the worst and that obtained by simultaneous exposure of moving image and noise was in between the two extreme.
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 the case where only the interior noise was exposed.
The height of the moving image also affected the subjective evaluation of car interior noise although the result was not shown here because of the shortage of the space.
Table 1 List of adjectives for SD experiment
FIGURE 1. Effect of moving scenery from the car inside to the perception of sound quality
a thin - thickb unsatisfactory - powerfulc booming - ringingd cheap - expensivee unpleasant - pleasantf clamorous - quietg dull - sharph muddy - cleari tight - loosej rough - smoothk hard - softl shrill - calmm heavy - light
1
2
3
4
5
6
7
a b c d e f g h i j k l m
eva
luat
ion
by imageby soundby image and sound
FIGURE 2. Effect of moving image from the car inside was equivalent to 10dB reduction in SPL.
EFFECTS OF SEAT/FLOOR VIBRATIONS AND SCENERY
In order to see the effect of seat/floor/steering wheel vibrations together with the moving scenery from the car inside, laboratory test using the test facility shown in Fig.3 was conducted. The vibrations were initially recorded in a real car at the positions of the seat, floor and steering wheel under the real running condition and these signals were reproduced in a sound proof 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 hearing.
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 responses was made using visual Basic language.
FIGURE 3. Test facility
The effect of simultaneous exposure of noise and seat/ floor/ steering wheel vibrations to the perception of unpleasantness and the effect of simultaneous exposure of noise and scenery were examined by comparing the result obtained only by noise exposure. As was seen from the Fig. 4, the effect of vibrations strengthened the unpleasantness and the effect of scenery weakened the unpleasantness.
FIGURE 4. Effect of vibration and scenery to the perception of unpleasantness of car interior noise
CONCLUSIONS
1. Effect of visual image reduces the negative impression of sound quality and the amount is sometime equivalent to 10dB reduction in SPL.
2. Seat/floor/steering wheel vibrations strengthen the unpleasantness while the scenery reduces the unpleasantness.
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 Forum, 1-4(2001), (in Japanese with English summary) .
0
1
2
3
4
5
6
7
abcdef ghi j kl m
eva
luat
ion
by soundby sound and imageby sound reduced 10dB
ProjectorProjectorProjectorProjector
ScreenScreenScreenScreen transducertransducertransducertransducer
SpeakerSpeakerSpeakerSpeaker WooferWooferWooferWoofer
SpringSpringSpringSpring ExciterExciterExciterExciter
seatseatseatseat
Steering wheelSteering wheelSteering wheelSteering wheel
headphoneheadphoneheadphoneheadphone
0
1
2
3
4
5
6
7
0 100 200 300 400 500
time (sec)
pleas
ant-
unpl
eas
ant
Noise
Noise+Vibration
Niose+Scenery
Inter-Modal Effects of Non-Simultaneous Stimulus PresentationM.E. Altinsoya, J. Blauerta ,C. Treierb
aInstitut für Kommunikationsakustik, Ruhr-Universität Bochum, D-44780 Bochum, GermanybInstitut 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 airplane, or when operating a machine or an appliance) non-acoustics factors like vibration, heat, visual effects may heavilyinfluence the judgements on product-sound quality. Auditory virtual-reality generators are potent tools for psycho-acousticresearch in complex, interactive auditory scenarios. They can be extended to serve as a tool for multi-modal psychophysics byincluding non-auditory modalities. Such multi-modal virtual-reality generators become more and more important as tools forproduct-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 withrespect to temporal factors, particularly synchrony. The objective of this paper is to provide an overview on this topic.
1. INTRODUCTION
In an earlier paper, we have defined product-soundquality as “a descriptor of the adequacy of the soundattached to a product. It results from judgements uponthe totality of auditory characteristics of the said sound- the judgements being performed with reference to theset of those desired features of the product which areapparent to the users in their actual cognitive, actionaland emotional situation” [1]. In user interaction withcomplex products various information reaches the userfrom different modalities. Consequently; the cross-modal information is of substantial influence when theevaluating of the product-sound quality. Auditory vir-tual-reality generators have proved to be potent toolsfor psychoacoustic research in complex, interactiveauditory scenarios and there are strong efforts recentlyto use them as car and aircraft simulators for product-sound-quality research [2,3,4]. One of the major ob-jectives of virtual-environment (VE) designers andresearchers is to obtain more realistic and compellingvirtual environments. This objective, though, requires abetter understanding of the integration of the main sen-sory modalities, namely, auditory, visual and tactile. Anunderstanding of perceptual aspects of temporal factorson multi-modal presentations is very important todetermine how to integrate multi-modal information.This paper presents some overview of temporal aspectsof multi-modal integration, based on literature data.
2. TEMPORAL ASPECTS OF MULTI-MODAL INTEGRATION
Multi-sensory integration has been defined by Kohl-rausch and van de Par as the synthesis of informationfrom two or more sensory modalities such that infor-mation emerges which could not have been obtainedfrom each of the sensory modalities seperately [5].Effects of non-synchronisation on perception (i.e..delays between modalities) are important factors for the
perceptual realism provided by VE Systems. Miner andCaudell’s computational analysis reveals that acousticprocessing delays of at least 66 ms must be expectedwith today’s technology for producing a rathersimplistic auditory field. As the complexity of auditoryenvironments increases, the computation time andresources will also do so. Audio delays of thismagnitude may have a negative impact on severalaspects of VE simulations [6]. Barfield et al. comparedthe human’s ability to detect and discriminate visual,auditory, tactile and kinesthetic information with thecurrent technical specifications of virtual-environmentequipment, but for each modalities seperately. Theirstudy does not include the human ability of multi-modalinformation detection and discrimination [7]. A multi-modal synchronisation threshold can be defined as themaximum tolerable temporal separation of the onset oftwo stimuli, one of which is presented to one sense andthe other to another sense, such that the accompanyingsensory objects are perceived as being synchronous. Inorder to measure this threshold, observers may be askedto report which of the two stimuli comes first (forcedchoice). In multi-modal-interaction research there areseveral studies regarding the detection ofsynchronisation thresholds. Most investigations whichdeal with the effect of temporal factors in multi-modalpresentation are related to audio-visual synchrony,while only very few investigations address auditory-tactile synchrony. The obtained results vary, dependingon the kind of stimuli and the psychometric methodsemployed. Hirsh and Sherrick measured thesynchronisation thresholds regarding visual, auditoryand tactile modalities [8]. They presented 666 Hz, 10-ms sine pulses as acoustic stimuli via headphone,similar pulses as tactile stimuli to the tip of the indexfinger via a shaker, as well as 5-ms flashes of light on ascreen as visual stimuli. The subjects were asked toreport which stimulus came first. Their results showthat the visual system is relatively sluggish while thetactual system is less so, and the auditory system seems
to be fastest (see Table 1). Dixon and Spitz investigatedsynchronisation-threshold differences between twodifferent stimuli, namely, speech and hammer-strikes[9]. They found that the synchronisation threshold ofspeech is higher than that for the impact stimuli. Minerand Caudell conducted a cross-modal psycho-acousticexperiment to measure perceptual
Table 1. Synchronisation-threshold values
Hirsh & Sherrickaudio visual audio delay visual delay
20 ms 20 msaudio-tactile audio delay tactile delay
25 ms 12 msvisual-tactile visual delay tactile delay
20 ms 30 msDixon & Spitzaudio-visual audio delay visual delay hammer-impact 188 ms 75 ms speech 258 ms 131 msMiner & Caudellaudio-visual audio delay hammer 177,17 ms drum Stick 176,70 ms glasses 191,15 ms Newton’s cradle 172,70 ms speech 203,32 ms overall mean 184,21 msLewkowiczaudio-visual audio delay visual delay adult 112 ms 65 ms infant 450 ms 350 msKohlrausch & van de Paraudio-visual audio delay visual delay AFC method 85 ms 29 ms constant. meth. 175 ms 75 ms
audio-visual-synchronisation thresholds for audio-signal delay [6]. Three different single impact events (adead-blow hammer striking a lead block, two wineglasses colliding, and two drumsticks striking), onerepeated-impact event (two suspended silver ballscolliding twelve times, Newton’s cradle), and speechsegment were selected to determine the thresholdvariation for different characteristic stimuli. Theexperimental results suggest a limited audio-processingbudget available in terms of audio-visual synchro-nisation requirements. Lewkowicz performed an ex-periment to investigate the AV-asynchrony-thresholddifferences between adults and infants. Participantswere familiarised with a bouncing disk and a sound thatoccurred each time the disk bounced. Then they weregiven a series of asynchrony test trials where the soundoccurred either before or after the disk bounced [10].Sensitivity to auditory-visual asynchrony was measuredby van de Par and Kohlrausch [11] using brief tonalsignals with a frequency of 500 Hz, accompanied by amoving disk on a monitor which was visible for 2000ms. In this study two different psychophysicalmeasuring methods (alternative forced choice, AFC,
and constant stimuli without feedback ) were used andobtained two different results. The authors report thatthe synchrony curves obtained are not centred aroundan AV delay of 0 ms. The reason for this may bedifferent time spans needed for the internal processingof auditory and visual stimuli. Obviously the humanperceptual system is adapted to tolerate large audiodelays which, consequently, may result in lowersensitivities to audio delays as compared to videodelays. Further, the experimental procedure chosen hasa strong influence on the results.
3. ONGOING RESEARCH
Many virtual auditory-tactile environments, e.g. carsimulators, require a whole-body tactile stimulation.For this case, literature data are not available.Therefore, further investigations on the auditory-tactilesynchrony are currently carried out by us to measure thesynchronisation threshold of auditory-tactile pres-entations using a whole-body vibrator and a virtualauditory environment. Both realistic and artificialstimuli are used. In the first case an auditory-tactilerecording was made of a car passing a bump. In thesecond case a broad-band noise for the auditorystimulus and a sine wave for the tactile stimulus wereemployed. During the experiments, tactile and auditorystimuli are presented with various inter-stimuli delays.The results of this investigation will be ready for oralpresentation at the congress.
4. REFERENCES
[1] Blauert, J., Jekosch, U., 1997. ACUSTICA/acta acustica, 83,747-753[2] Genuit, K., 2000,.“The future of sound quality of the interiornoise of vehicles”, in: Proc. Internoise 2000, 1693-1698, F-Nice,[3] Quehl, J., Schick, A., Mellert, V., Schulte-Fortkamp, B., Rem-mers, H. 2000. “Dimensions of Combined Acoustic and VibrationPerception in Aircrafts Derived by Factor Analyis of SemanticDifferential Data”, in Proc. Internoise 2000, 465-469, F-Nice[4] Hillebrand, P., Schaaf, K., 2000. “Anwendungen eines vibroa-kustischen Simulators in der Automobilindustrie”, in Fortschr.Akustik, DAGA 2001, Dtsch. Ges. Akustik, D-Oldenburg[5] Kohlrausch A., van de Par S., 1999. “Auditory-visual interac-tion: From fundamental research in cognitive psychology to(possible) applications”, in: Human Vision and Electronic ImagingIV, Proc. Soc. Photo-Optical Instrumentation Engrs. 3644 , 34-44[6] Miner N., Caudell T., 1998. Presence 7, 396-409 (1998)[7] Barfield, W., Hendrix, C., Bjorneseth, O., Kaczmarek, K.A.,Lotens, W., 1995. Presence 4, 329-356[8] Hirsh I.J., and Sherrrick C.E, 1961. J. Exp. Psychol 62, 423-432Dixon N.F., Spitz L., Perception 9, 719-721[10] Lewkowicz, DJ, 1996. J. Exp. Psych. 22, 1094-1106[11] Van de Par S., Kohlrausch A., 1999. IPO Ann. Progr. Rep.34,94-102
Psychoacoustic Correlates of Time and SpectralCharacteristics 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 velocity and the distance of the passingtrain. Annoyance judgments of a moving train cannot be merely explained on the basis of the spectral content and sound levelof noise source. Loudness spectra and time patterns of the noise of a train moving at the same velocity but recorded at differentdistances deliver important cues that help to account for the differences in annoyance judgements. The autocorrelation functioncalculated on the time patterns of all stimuli allows to identify periodic components occurring in the stimulus. The paper aimsto determine to what extent spectral and time characteristics of railway noise contribute to its annoyance.
INTRODUCTION
Noise produced by source located farther from alistener is assessed as less annoying than noiseproduced by source of the same type that is locatedcloser. This is explained by pointing to the fact thatincrease in distance of the source of noise causesdecrease in loudness. However, the ground effect andair absorption phenomenon change not only the soundlevel of the stimulus but also its spectral content. The aim of the present paper is to answer thequestions: (1) what is the annoyance of noise producedby similar sources located at different distances butreaching listener's ear with the same loudness? (2) towhat extent spectral and time characteristics of railwaynoise contribute to its annoyance? In the present work, noise of a train moving atdifferent velocities, recorded at two distances was theobject of annoyance judgments. The original 12railway noises recorded at two distances wereartificially modified to make them equal in loudness.The loudness equalization was done in two steps. Atfirst, the sound recorded at the farther distance wasamplified in a linear way until its LAE was the same asthe LAE of the noise recorded at the closer distancefrom the source. Then, based on the spectra of thesenoises the loudness, N, according to the Zwickermethod [3] was (ISO532B) calculated. This procedureguaranteed that the noises presented in pairs werealways equal in LAE but not always in their loudness.In the psychoacoustic experiment, subjects were askedwhich of the two noises presented in a pair they wouldprefer to switch off given such a possibility.
Stimuli and apparatus
Twelve original railway noises, recorded at twodistances (S1=25m and S2=450m) from the moving
source were used as test stimuli. There were noisesgenerated by Inter City (IC), passenger (PT) andgoods trains (GT), each of 25s duration. The velocityof each train is presented in Table 1.
Subjects
Subjects were normal-hearing students of theA. Mickiewicz University (15 female and 16 male).
Procedure
Subjects were given the following instruction: “Youwill listen to a pair of railway noises, please markwhich of the two noises presented in a pair you wouldprefer to switch off, if given the possibility”. Eachsubject judged each pair of noises only once. Theresults are presented as the percentage of occurrencesof a particular noise chosen as non-preferredcomponent in a pair to all occurrences of this noise.The higher this non-preference percentage, the greaterwas the number of subjects who wanted to switch thisparticular noise off. In all, each of the 31 subjectsmade 24 preference judgements, one for each of 24pairs.
RESULTS
The results of these comparisons are presented in theTable 1. The pairs of noises equal in loudness aremarked in the Table 1 with an asterisk. We assumethat results above 75% indicate that subjects’ choiceswere not casual. It can be seen that subjects’ choices ofthe more annoying component are in agreement withthe lower frequency of the periodic component foundin the autocorrelation function calculated for all noises.Periodic component is defined by the time delay at the
first maximum peak of the normalized autocorrelationfunction [1] (it corresponds to the phenomenon of thevirtual pitch [2]). It seems, that periodic components ofnoise generated by IC train were too high to favor ofthe choice of the more annoying component.
FIGURE 2. Averaged spectra of L AE, for original and
modified PT train noises.
Table 1. Percentage of occurrences of noise chosen as non-preferred component in a pair (column s 2and 3), frequency of periodic component (columns 4 and 5), velocity of the train (column 6).
Train S1-org[%] S2-mod[%] S1-orgPeriod.
Comp.[Hz]
S2-modPeriod. comp.
[Hz]
Velocity, V[km/h]
*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
FIGURE 1. Averaged loudness spectra for original and
modified PT train noises .
Low frequency components occurred in original noisegenerated by PT trains. Noises with these lowfrequency components were chosen as distinctivelymore annoying. These low frequency components canbe seen in loudness spectra in Figure1. However, theyare not visible, when we present the same stimuli asthe averaged sound exposure spectra expressed in dBA(see Figure 2).
The differences in annoyance judgments of IC trainnoise compared to the PT and GT train noise may beexplained by pointing to the higher velocity of ICtrains. This difference might be the result of the
Doppler effect. If it is true, then annoyance judgementsof noise produced by the moving source depend onsource’s velocity.
REFERENCES
1. Y. Ando, H. Sakai and S. Sato., Journal of Sound andVibration 232, 101-127 (2000).
2. E. Terhardt., Hearing Research 1, 155-182 (1979).
3. E. Zwicker and H. Fastl., Psychoacoustics-Facts andModels, Heidelberg; Springer-Verlag, 1990, pp. 289-291.
0,010,020,030,040,050,060,070,080,090,0
50 100
200
400
800
1600
3150
6300
Frequency [Hz]
Soun
d ex
posu
re le
vel [
d
PT-ORGPT-MOD
0,000,501,001,502,002,503,003,504,004,50
0,5 3,5 6,5 9,5 12,5
15,5
18,5
21,5
Barks
Spec
ific
loud
ness
[son
e/ba
r
PT-ORGPT-MOD
Onomatopoeic Features of Sounds Emitted from Laser Printersand 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, JAPAN2 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 parametersin the onomatopoeic representations among sound stimuli were applied to the hierarchical cluster analysis. As a result, thesound stimuli were categorized into 5 clusters, according to the acoustical properties in the frequency domain and the timedomain. Furthermore, the product images associated with its sounds were measured by a rating experiment. Severalrelationships between the product image and phonetic features were clarified: for example, the prolonged sound of /i/ andvoiced consonants are associated with unpleasantness.
INTRODUCTION
Recently, in offices, there are many kinds of soundsfrom office equipment, such as computers, printers andcopy machines. Printers and copy machines emit varioustypes of manual operation sounds when we open andclose top covers, draw and load paper trays, and lock andrelease hooks. These sounds contribute to the quality ofproducts and the reliability of operation. Users often feelanxiety from some types of manual operation sounds. When users complain about manual operation soundsto distributors, onomatopoeic representations arefrequently used. The onomatopoeias are a natural way toexpress auditory sensations. They may reflect the acousticfeatures which affects the auditory impression andimagery. The onomatopoeic representations can be usedto measure product quality without actually doing theformal psychoacoustical experiment. The psychoacoustical experiments were conducted toinvestigate the possibility of using onomatopoeic repre-sentations to estimate the quality of laser printers andcopy machines. Firstly, the free description evaluationexperiment using onomatopoeic representations wasexamined for manual operation sounds. Furthermore,subjective evaluation experiments using adjective scaleson the product image were examined. Then, therelationship between the product image and the onomato-poeic features was discussed.
EXPERIMENTAL METHOD
Forty four manual operation sounds, such as openingand closing front covers, drawing and loading paper trays,and locking and releasing hooks of 3 laser printers and 2copy machines were used as the sound stimuli. Most ofthem have impulsive striking sounds emitted from theoperated parts such as front covers, paper trays and hooks.They were recorded in an office-like room using adummy head recording system. These stimuli werepresented to subjects from a computer via headphones. Two kinds of experiments were conducted. One was
the free description experiment using onomatopoeicrepresentations for the manual operation sounds. Theother was the measurement of the impression of the samesound stimuli using a 7-step adjective scale of “pleasant –unpleasant” and “strong – breakable”. Fourteen nativespeakers of Japanese (6 male and 8 female subjects)participated in these experiments.
DISCUSSION
Analysis of Onomatopoeic Representations To clarify the onomatopoeic features to represent theacoustical properties of the manual operation sounds, theonomatopoeic representations obtained were coded using24 phonetic parameters such as 7 places of articulation, 6manners of articulation [1], 5 vowels in Japanese, voicedand voiceless consonants, syllabic nasal, choked sound,palatalized sound and prolonged sound. Furthermore,cluster analysis was applied to the similarities among thesound stimuli expressed by the frequencies of the 24phonetic parameters. The result of cluster analysis isshown in Figure 1. The type of stimuli, examples ofonomatopoeic representations and the average number ofphonemes are shown in each cluster. All stimuli werecategorized into 5 clusters. The sound stimuli in cluster 1 and cluster 2 consist ofrubbing sounds from the operated part and impulsivestriking sounds. The rubbing sounds in cluster 2 haveespecially high energy in the high frequency region. Inonomatopoeic representations for these “sharp” sounds,the Japanese vowel of /i/ [i] and its prolonged sounds arefrequently used such as /kiiiii/ [k i ] (/keeey/ in Englishexpression). The Japanese vowel of /i/ [i] with the highestsecond formant frequency (about 2.8kHz) in the vowels isused in order to represent the sharp impression [2]. Therubbing sounds in cluster 1 have energy not only in thehigh frequency region but also in the lower frequencyregion. In the onomatopoeic representations, the othervowels, except /i/ and their prolonged sounds, are oftenused such as /shuuuu/ [ ].
FIGURE 1. Dendorogram of the manual operation sounds
The sound stimuli in cluster 3 have only striking sounds.Therefore, the onomatopoeic representations for them areshorter than those in the other clusters. The averagenumber of phonemes in the onomatopoeic representationsis 3.5. Examples of the onomatopoeic representations forthese stimuli are simple, such as /ban/ [ba ], /kan/ [k ]and /dan/ [da ]. Two sound stimuli categorized in cluster 4 are the long-est in duration. They have continuous sounds emitted fromthe operated parts after striking. Therefore, the onomat-opoeic representations for them are longer than those inother clusters. The average number of phonemes in theonomatopoeic representations is 8.7. As an example of theonomatopoeic representations for the continuous sounds,repeated syllables are used such as /ton ton ton/[tontonto ]. The sound stimuli in cluster 5 have many short soundssuccessively emitted from the operated parts beforestriking. There are short pauses between many shortsounds. To represent the short sounds with the shortpauses, the stop and the flapped articulated in the alveolarare often used such as /ka ta ta ta/ [k tatata]. From these results, the onomatopoeia is a valid way tocapture the acoustical properties of the manual operationsounds in the time and frequency domains.
The Relationships between Product Imageand Onomatopoeic Features To clarify the relationships between the product image,such as pleasantness and strength associated with soundstimuli, and the onomatopoeic features, the number ofwhole phonemes in the onomatopoeic representations ofall subjects for each sound were counted in each period ofthe striking sound, the previous sound and the following
sound. Furthermore, the rank correlation coefficients be-tween the number of each phoneme and average pleasant-ness scores were calculated. In the group of unpleasant sounds, there are all stimulihaving the “sharp” rubbing sound before the strikingsound from the operated part, as shown in Figure 2. It issupposed that these stimuli were evaluated to be un-pleasant due to the “sharp” rubbing sounds. These sharpand unpleasant sounds are represented by consonants withthe vowel of /i/ and its prolonged sounds such as /kiiiii/[k i ] and /giiiii/ [ i ] (/geee/ in English expression). Thecorrelation coefficients between the average pleasantnessscores and the number of phonemes /i/ [i], /hi/ [ i] and/kyu/ [k j ] are statistically significant at a level of 0.01(r=0.439, r=0.417 and r=0.445 ). The beginnings of the onomatopoeic representations forthe impulsive striking sounds emitted from the operatedparts are expressed by the stops, such as /ban/ [ba ], /pan/[pa ] and /ga tan/ [ ta ]. Especially, for the un-pleasant sounds, the stops of voiced consonants such as /g/and /b/ (for example, /ga/ [ ], /ba/ [ba] and /be/ [be]) areused at the beginnings of the striking sounds morefrequently than those for the pleasant sounds. The spectraof the voiced consonants have higher energy in the highfrequency region than that of the voiceless consonants [3].Therefore, to represent the “harsh” striking sounds withhigh energy in the high frequency region, the voiced con-sonants should be used. The correlation coefficientsbetween the average scores on pleasantness and thenumber of phonemes such as /ga/ [ ], /be/ [be] and /gwa/[ a] are statistically significant at a level of 0.01(r=0.406, r=0.453 and r=0.393). The correlationcoefficients between the average scores in the image ofstrength and the number of phonemes such as /ga/ and/be/ also are statistically significant.
FIGURE 2. Wavelet analysis of the most unpleasant soundwith a “sharp” rubbing sound and a “harsh” striking sound
REFERENCES1. H.Jyouo, “Phonetics in Japanese”, Bandai Music Entertainment, Tokyo,1998 (in Japanese)2. S.Iwamiya and M.Nakagawa, “Classification of audio signals usingonomatopoeia”, Soundscape, Vol.2, 23-30 (2000) (in Japanese)3. K.Tanaka, K.Matsubara and T.Sato, “Onomatopoeia expression forstrange noise of machines”, J. Acous. Soc. Japan, Vol.53 (6), 477-482(1997) (in Japanese)
1: locks, trays, covers[s ba i ]，[ bata ]phonemes: 5.4
2: covers[k i ba ]，[k j i t a ]phonemes: 6.0
3: covers[ba ], [k ], [da ]phonemes: 3.5
4: trays[do k a a ado tontontonto ]phonemes: 8.7
5: covers, sliders, locks[ a a a ta ]，[k tatata ta ]phonemes: 5.0
The number of cluster：type of stimulusExamples of onomatopoeic representationsAverage number of phonemes
0.010.02
0.050. 10.2
0.512
5
1020
[kHz]Time [s]
Sharp rubbing soundSharp rubbing soundSharp rubbing soundSharp rubbing sound
Harsh striking soundHarsh striking soundHarsh striking soundHarsh striking sound
222222220000000000000000mmmmmmmmssssssss
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 fundamental frequency 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
We have neurophysiologically and behaviorally investigated pitch extraction system in the Japanese monkey [1,2,3]. Our behavioral studies have revealed that Japanese monkeys show similar pitch perception that humans do including the missing fundamental perception[2]. In this research perception of directional pitch change in monkeys and humans are compared.
METHODS
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+)
was presented, they could drink water from the spout. Pushing the button during negative stimulus (S-) was punished with a time out. Sound stimuli were presented from a loud speaker fixed in front of the animal. Stimuli were sequentially presented simple tone bursts and harmonically structured complex tone bursts (Figure 2). Figure 2a indicated envelope pattern of two sequential tones. For Subject M1, S+ was defined by an increment of the fundamental frequency between the two tones while S- was defined by the opposite direction. Schematic spectrograms of a set of
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.
tones for Subject M1 were indicated in Figure 3. For Subject M2, S+ and S- were switched. T1 was a pair of tone bursts. T2 and T3 were pairs of complex tones. T4 and T5 were combination of a tone burst and a complex tone. The frequency of the simple tone burst was in between the fundamental frequency and the second harmonic of the complex tone burst for T4. The fundamental frequency of the complex tone burst was higher than the frequency of the simple tone burst for T5. T6 and T7 were a pair of tone bursts where the highest harmonics were the same frequency. The fundamental frequency was selected randomly from 150 – 1000 Hz. The frequencies of the highest harmonics were lower than 9000 Hz. The frequency ratio between the first and the second tones in a sequence was 1.25 - 1.89. Sound pressure level was randomly varied between 30 and 50 dB SPL. The stimulus set was replaced with an unfamiliar set each session. In order to compare data obtained from humans, seven human subjects were used in the same paradigm except of the water reward. For human 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 1000 ms (Figure 2a), or interval between two tones was varied from 0 to 800 ms (Figure 2b).
RESULTS AND DISCUSSION
The monkeys were successfully trained to discriminate the direction of fundamental frequency and/or pitch change, rising or falling in the same manner as our precedent experiment with the correct ratio above 90 % [4] except for the tone pattern T4 [5]. For T4 pattern, correct ratio of Subject M1 was 70 % and that of Subject M2 was 80 %, and they tended to respond reversely, in other words, they pushed button during S- and not during S+ especially for
fundamental frequency lower than 300 Hz persistently. 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 with the change in the fundamental frequency. Others often showed improvement as tone duration lengthened or interval between two tones lengthened (Figure 4).
Positive stimuli (S+) Negative stimuli (S-)
Freq
uenc
y
Time FIGURE 3. Schematic spectrograms of a set of tones for Subject M1.
���������������
�������������������������
���������������
������������������������������
������������������������������
������������������������������
�����������������������������������
����������������������������������������
�������������������������
����������������������������������������
����������������������������������������
�����������������������������������
���������������
�������������������������
�������������������������
�������������������������
0
1
2
3
4
5
6
0 200 400 800
interval (ms)
responses
����S1����S2����
����S3��������S4
FIGURE 4. Number of correct responses at each interval between tones for human subjects.
ACKNOWLEDGMENTS
This research was supported by Special Coordination Funds from the Science and Technology Agency and a grant to RCAST at Doshisha University from the Ministry of Education and Science of Japan.
REFERENCES
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.
2. K. Manabe, S. Toriyama, and H. Riquimaroux, Proc. Autumn Meet. Acoust. Soc. Jpn. 457-458 (1997)
3. N. Kitagawa, S. Toriyama, and H. Riquimaroux, Proc. Spring Meet. Acoust. Soc. Jpn. .381-382 (2000)
4. N. Kitagawa and H. Riquimaroux, Trans. Tech. Comm.
Psychol. Physiol. Acoust. , Vol. H-2000-31,1-8 (2000)
5. T. Takahashi, K. Sumida, Y. Yanase, and H. Riquimaroux, 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
This paper describes the results of the time varying evaluation on sound quality of noise emitted from a hydraulic powered earth-moving machine. From the viewpoint of an operator’s comfort and that of the inhabitants living near the construction site, exterior as well as interior noise was used together with video movies showing the interior and exterior scenery of the machine in operation for evaluation. The time variation of several sound quality parameters such as loudness, sharpness, roughness, fluctuation strength and booming index[1] were calculated with appropriate time constant, i.e., 1.7 seconds for better correlation with the subjective responses. A regression model in terms of these parameters was used for estimation of unpleasantness. As a result, good correlation was obtained between the evaluation of unpleasantness and the estimation obtained by the model.
EXPERIMENT
The machine repeats, “dig-turn-throw-turn” four times and exterior noise at four locations was recorded, i.e., at left front, at left back, at right front and at right back and the distance from the machine to the four locations was 10 m. Interior noise was recorded with the conditions where an air conditioner switched on and off with interior video scenery. The sound stimuli were presented to the subject through an electro-static headphone inside the sound proof room. The subject
FIGURE 1. Evaluation of exterior noise on un-pleasantness and time variation of loudness
（left-front）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90（ｓ）
ple
asan
t
unple
asan
t
20
25
30
35
40
45
50
loudness
（so
ne）
bad
good
（left-back）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90
（ｓ）
ple
asan
t
unple
asan
t
20
25
30
35
40
45
50
loudness
（so
ne）
bad
good
（right-front）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90
（ｓ）
pleas
ant
unpl
eas
ant
20
25
30
35
40
45
50
loudness
（so
ne）
bad
good
（right-back）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90
（ｓ）
ple
asan
t u
nple
asan
t
20
25
30
35
40
45
50
loudn
ess
（so
ne）
total impression for sound with scenery total impression for sound onlysound with scenery sound onlyloudness
bad
good
ｒ＝０．３８７（sound with scenery）ｒ＝０．３８１（sound only）
ｒ＝０．６６９（sound with scenery）ｒ＝０．６３２（sound only）
ｒ＝０．７８５（sound with scenery）ｒ＝０．７９５（sound only）
ｒ＝－０．０３６（sound with scenery）ｒ＝－０．０４６（sound only）
was asked to put his/her instantaneous response on unpleasantness, powerfulness, sharpness and booming sensation into a personal computer using a keyboard as an input device for about 80 seconds from the beginning to the end. After the evaluation on unpleasantness subjects were asked to answer their overall impression on sound. 14 males and 1 female for unpleasantness and powerfulness and 22 males and 1 female for sharpness and booming sensation joined as subjects. They all had normal hearing and were aged between 22 to 54 years. Their individual responses in each four factors were averaged over total subjects to get the final results.
RESULTS
The results on unpleasantness at the four exterior locations were shown with the variation of loudness in Figure 1. The responses varied with time according to the variation of loudness and sharpness. The results obtained by presenting sound only were more unpleasant, more powerful and sharper than those obtained by presenting sound with scenery. About total impression, evaluation at the right back was better than the one obtained at the right front though loudness at the right back was larger than that at the right front. This result was due to the fact that large variation of loudness influenced the total impression.
REGRESSION MODELS
We calculated objective measures such as loudness, sharpness and booming index with a time constant of 1.7 seconds. These were used in the regression analysis for rating unpleasantness. The model for unpleasantness for exterior noise with scenery was
FIGURE 2. Evaluation of interior noise on unpleasantness and the variation of booming index
obtained utilizing sharpness and the slope of sharpness between 8 seconds as explanatory variables and the one for exterior noise only was obtained utilizing sharpness and the slope of loudness between 10 seconds as shown in Figure 3. Models for unpleasantness for interior noise were obtained utilizing loudness and booming index as explanatory variables, but due to the shortage of space the results were not shown here.
CONCLUSIONS
1. The results obtained by presenting sound only were more unpleasant, more powerful and sharper than those obtained by presenting sound with scenery.
2. The evaluation of sound quality of exterior earth-moving machine noise was influenced by its loudness and sharpness and the one of interior noise was influenced by its booming index and loudness.
FIGURE 3. Evaluation of exterior noise on unpleasantness and the estimation by a regression model in terms of sharpness and slope of loudness in 10 seconds in case of sound only
REFERENCE
1. S. Hatano and T. Hashimoto., Proceedindings Inter-Noise 2000, No.233, Nice, 2000, pp. 1-4
図 車内音不快感評価と音圧レベル
（with air condisioner）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90（ｓ）
pleas
ant
unpl
eas
ant
20
25
30
35
40
45
50
boom
ing
index
bad
good
（without air condisioner）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90
（ｓ）
pleas
ant
unpl
eas
ant
20
25
30
35
40
45
50
boom
ing
index
total impression for sound with scenery total impression for sound onlysound with scenery sound onlyloudness
bad
good
ｒ＝０．４３５（sound with scenery）ｒ＝０．４１４（sound only）
ｒ＝０．４０１（sound with scenery）ｒ＝０．３７２（sound only）
（ le ft-front）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90（ｓ）
ple
asan
t
unple
asan
t
（ left-back）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90（ｓ）
ple
asan
t
unple
asan
t
（right-front）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90（ｓ）
ple
asan
t
unple
asan
t
（right-back）
1
2
3
4
5
6
7
0 10 20 30 40 50 60 70 80 90（ｓ）
ple
asan
t
unple
asan
t
sub jective evaluation regression estimateR=0.900R=0.900R=0.900R=0.900
Aspects on three methods for paired comparison listening testsA-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 ignoredand 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
It is sometimes hard for a subject to use and for anexperimenter to analyse results using scales in the evalu-ation. There are uncertainties whether the subjects haveused and understood the scale equally. The problem isavoided using paired comparison tests where subjectsare asked to judge which of two treatments has a cer-tain attribute (e.g. a pleasant sound). This paper givessome aspects on some of these models. Two majorpair comparison methods exist today, the so-called Thur-stone model [9] and the Bradley-Terry model (BT-model)[1, 2]. They give similar results but since there exist more(comprehensive) extensions to the latter, the focus in thispaper will be set on the BT-model and its extensions onties [4, 8].
METHODS
Test without ties
The BT-model is a model that gives maximum-likelihood estimates of the treatment ratings,Ti , i =1, ..., t. The probability of choosingTi when compared toTj is given in eq. (1) whereπi , ...,πt represent relative se-lection properties for thet treatments,i 6= j, i, j = 1, ..., t.1
P(Ti→Tj)=πi
πi +π j=Z ∞
−(lnπi−lnπ j )sech2(y/2)dy (1)
The probability can be described as an integral as shownabove, where the probability is seen to be dependent onthe natural logarithm of theπ-values of the treatments.This is the reason why comparison of the treatmentsshould be made on the natural logarithm of the preferencevalues, called the true merits. Using a generalization of abinomial model and its distribution, the complete likeli-hood function,L, becomes
L = ∏i< j
(
ni j
ai j
)
∏ti=1 πai
i
∏i< j(πi +π j)ni j(2)
1 In all BT- models, comparisonsTi −Ti are not allowed.
whereai j is the number of timesi was selected anda ji
is the number of timesj was selected(ai j + a ji = ni j ).By maximizing the natural logarithm ofL and using theconstraint∑t
i πi = 1, estimates ofπi are obtained.2
If the objective of the listening test is to �nd out if alltreatments (or sounds) are perceived similar or not, hy-pothesis tests presented in [1, 2] can be used. If the ob-jective, on the other hand, is to receive a ranking of thetreatments and to say that one of the treatments is betterthan another treatment on a certain signi�cance level,3 adifferent approach is needed. When choosing an appro-priate approach it is important to notice that the estimatesof πi are not independent and that the variances of eachtreatment are not necessarily homogeneous. It is there-fore not appropriate to use ANOVA-tests. We can insteadcalculate the variance and covariances for the treatmentsas described in [2]. They form an ellipsoidal region4 asshown in Figure 1 and its area is dependent on the chosensigni�cance level. The region tells us that the estimatesof πi andπ j should exist somewhere within that area. Ahypothesis test of any point outside the region will be re-jected on the chosen signi�cance level, and if the regionis crossed by the plane of symmetry the hypothesis thatthey are similar cannot be rejected. For the two differ-ent signi�cance levels in Figure 1 it would be correct tosay the treatments are different for the smaller area withthe higherα, but not for the larger area with the lowersigni�cance level.
2 Problems may arise when one or more of the treatments always arechosen in favour of the others.3 The signi�cance level,α is the probability to reject a hypothesis eventhough it is true. Loweringα sets a stronger demand on the hypothesis.4 By looking at the region we can check if our estimates are indepen-dent and the variances homogeneous. If the region forms a circle, thevariance is equal for the estimates and the estimates are independent.If an ellipse, nonparallel to any of the axis, is given, as in Figure 1,the estimates are not independent. Furthermore, if it is not parallel tothe symmetric plane, the variances of the estimates are inhomogeneous.Plots similar to Figure 1 were given when a listening test was performedat the division on various drum sounds from �oor coverings. The useof ANOVA-tests would therefore not be correct. A re�nement of theprocedure can be made using contrasts [7], but this procedure is notincluded in this paper.
0 10
1
treatment i
trea
tmen
t j
FIGURE 1. Con�dence regions for twoα, comparing treatmenti and j. The smaller area is given by a higher value ofα.
Test with ties but not used in the analysis
This type of test was introduced when no model forhandling ties existed. The random answers for compar-isons where subjects cannot make a difference are elimi-nated most simply by allowing them in the test but ignor-ing them in the analysis. Some experimenters divide theties by splitting them equally on the treatments. In [5]a test made by Hemelrijk is described where it is provedthat leaving ties out of consideration makes a more pow-erful test than if the ties are equally distributed. Today,however, models capable of handling ties are available,and this procedure is therefore not recommended, as in-formation is lost.
Test and analysis with ties
The idea behind the Rao and Kupper model (RK-model) [8] is that when the difference between two treat-ments is smaller than a certain value, or threshold, thesubjects will declare a tie. The probability of choosingTi
when compared toTj is therefore (cf. eq.(1)) set to
P(Ti→Tj)=Z ∞
−(lnπi−lnπ j )+ηsech2(y/2)dy=
πi
πi +θπ j(3)
whereη=ln(θ) is the sensory threshold for the subject.The probabilities for preference ofj or for a tie is calcu-lated using the integral and are given in [8].
Davidson presented another extension to handle tiesbased on the BT-model. His approach was to ensure thatan axiom of choice presented in [6] is ful�lled. Theidea is that alternatives which should be irrelevant to thechoice are in fact irrelevant, stated asP(i|i, j)/P( j|i, j) =πi/π j , which the RK-model does not ful�l. It is assumedthatP(0|i, j) = ν
√
P(i|i, j)P( j|i, j), whereν is seen as anindex of discrimination. The assumption of a geometricmean are based on the fact that the merits, ln(π), can be
represented on a linear scale. The probability for prefer-ence ofi when presented withj then becomes
P(Ti→Tj)=πi
πi+π j+ν√πiπ j;P(Ti=Tj)=
ν√πiπ j
πi+π j+ν√πiπ j
The estimates in the two models and their variancesand covariances are obtained as in the BT-model, us-ing maximum likelihood functions, and are described intheir articles. It has been noticed [2, 4] that the modelsare asymptotically equal and the choice of method is amatter of which idea seems more appealing. David [3]points out that the ful�lment of the choice axiom mightnot be required. A difference is, however, that the modelby Davidson gives a ranking that is always consistent toa ranking common in sports where a win is awarded 2points and a tie 1 point, which the RK-model is not.
REMARKS
When allowing ties, subjects might declare a tie al-though they, with some effort, could detect a difference.An investigation [5] on this problem gave the followingrecommendations. When discrimination is the objectiveit is better to prohibit ties as the subjects' ef�ciency ofdecision might be offset, but when preference is the ob-jective, ties should be allowed as they add information. Ithas been noticed by the present author that subject prefer-ences are not always normally distributed. Some subjectsprefer a darker drum sound while others prefer a higherpitched sound. If ties are prohibited a 50-50 relationshipof two treatments would indicate equal treatments, but ifties are allowed, keeping the same relationship and no tiesis reported, we can suspect the treatments are not equalbut only given the same amount of preferences.
REFERENCES
1. R. A. Bradley and M. E. Terry,Biometrika39, 324-345(1952).
2. R. A. Bradley,Handbook of Statistics4, 299-326 (1984).
3. H. A. David, The method of paired comparisons, 2nd Ed.,London: Grif�n, 1988.
4. R. R. Davidson,J. Amer. Stat. Assoc.65, 317-328, (1970).
5. N.T. Gridgeman,Biometrics15, 382-388, (1959).
6. R.D Luce, Individual Choice BehaviorNew York:Wiley,1959.
7. D. C. Montgomery,Design and Analysis of experiments5thEd., New York: John Wiley and Sons, 2001.
8. P. V. Rao and L. L. Kupper,J. Amer. Stat. Assoc.62, 194-204, (1967). Corrigenda,631550.
9. L. L. Thurstone,Amer. J. Psychol.38, 368-389, (1927).
Product Acoustics: Designing the Sound Qualities of aManufactured World
L. FuksEscola 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. Wheneverthese "by-product" sounds are considered as undesirable, the obvious procedure by acousticians is to employ methods formaximally suppressing them. However, those sounds may in many cases be useful to provide information about the properbehavior of the system. Examples are the sounds produced by car and motorcycle engines, "clicking" sounds from photo camerasand safety locks, "crispy" sounds of cream crackers, "buzzing"sounds of a microwave apparatus, and so many others. Thesesounds, more than a mere monitor of electro-mechanical performance, may give some kind of aesthetical satisfaction and alsoconvey 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 someguidelines for its establishment.
INTRODUCTION
It seems appropriate to bring out this discussion inItaly, where Marinetti founded the Futurist movementaround 1909. Luigi Russolo launched "L'arte deiRumori" futurist manifesto in 1913, proposing a newattitude towards the sonic ambient brought bytechnology. Russolo and co-workers created a newmusical style, a notation system and a set of twenty-seven instruments, the “noise intoners” (intonarumori). The sounds and noises produced by machines andother products are frequently regarded as undesirable,annoying or irrelevant. Yet, one could hardly think of"mute" devices working into complete silence. The“voices” of products convey rich and usefulinformation [1]. Nevertheless, the sonic outcome of the product isoften an uncontrolled, random and overlookeddimension in industrial design and engineering. Fromthe user's perspective, interaction with the productincludes the auditory stimuli produced, which mayprovide clues for the identification and control of itsfunction and even serve as a source of aesthetic delight.Furthermore, the sound from an object is in itself anobject [2]. Thus, it seems valid to gather knowledge onthe nature and on the aesthetical impact of thesestimuli. Let's focus on a photo camera, which is expected toproduce the classic click sound whenever a picture istaken. The qualities of this click may indicate whetherit is a professional or just a cheap disposable model.However, in situations requiring silence, the cameraturns into a disturbing noise source. Several issues
arise: (i) how the chain of mechanisms and interactionsresult in a particular click?; (ii) how does a goodcamera sound like?; (iii) how can the sound bemodelled, improved and possibly applied to a "badsounding" camera?; (iv) can the click be muffled orsuppressed when required?; (v) should a digital cameraclick the classic click?; (vi) should standards beestablished for product sounds? To our knowledge, no present discipline offersresources for dealing with such problems, in spite ofmost of them being related to acoustics andpsychoacoustics. Based on the issues above, we willdraft a systematic approach towards a proposeddiscipline, Product Acoustics.
VIBRO-ACOUSTICAL ANALYSIS
There is a need of a thorough analysis of thematerials employed, the types of mechanisms,accelerations and forces involved, the friction andshocks that take place and the way how the vibrationsare converted into sounds. This approach combinescomputational solid mechanics, acoustics andvibration, dynamics of mechanisms, science ofmaterials, among others. This aspect will considerablybenefit from knowledge on the acoustics of percussioninstruments (idiophones).
AUDITORY SEMIOLOGY
Users seem to associate product qualities to a hugedatabank of memorized sounds, which may result in theimpression of robustness, lightness, "nobility",
durability, etc. On the other hand, the sounds maydenounce malfunction, lack of adjustment, the presenceof cracks in the structure, among others. Therefore, anextense library of sounds must be acquired andevaluated by subjects through systematicpsychoacoustic and objective procedures.
SOUND RECORDING, ANALYSIS,REPRESENTATION, MODELLING
The sound signals produced by products may beroughly divided into transient, chaotic and periodic.The click of the camera is an example of a transientsound [see 3]. Each type requires appropriate methodsfor recording, analysis and representation [see 4]. Inthis aspect, a multidisciplinary approach combining theareas of signal processing, music acoustics, chaostheory, among others, is required.
AURAL COMMUNICATION/ ETHOS
Similarly to visual comunication, auralcommunication impregnates the product with ambienceand useful language. Here the sounds are "telling" thatthe camera is about to take the picture, that the picturehas been taken, that the battery charge is low, etc. Forthe user's convenience, these signs might be customizedto be pleasant and more easily recognized, particularlyin a very rich "soundscape" typical of present times [5].Also, these functions must comply with behaviouralconstraints (ethos), so that it does not disturb otherfunctions and individuals. The user should be able toturn the sound functions off and to reduce the outputlevel. Besides, sounding functions represent a relevantfactor in power consumption of batteries.
SOUND MODIFICATION/ MIMICKING
The designer should now be able to intervene on theclick sounds. This may consist of modifications in themechanism, materials, superficial texture anddimensions so that the sounds will be improved to meetsome expected results. Mechanical and electronicressonators and filters may be incorporated to alter theoriginal sound. Experience from the area of sounddesign will be called for in this aspect [6]. In order tomimic the sound of a desired click, a sound generatorusing digital synthesis may be added. This would be thecase of sounds aggregated to a digital camera which, inprinciple, could work in complete silence. A commoncomplain to silent cameras is that the photographer isnot always sure if the picture has been taken. The most common, and frequently disturbing, case ofsound mimicking is that of the "ring" of cellular
phones. Makers aggregate fragments of musical worksand various patterns, apparently without criteria orcompliance with any social rules or industrial standards
STANDARDIZATIONAs proposed above, product acoustics is
directly related to sound and noise emission, to thequality of the product and to sound communication andsignalling. There are several ISO standards that refer tonoise emission but they are mostly based on maximaladmitted loudness levels. ISO-9000 series refers tototal quality that contemplates several aspects ofproduct acoustics, such as interface with the user,subjective aesthetical characteristics and convenienceof use. However, to our knowledge, no standards referto sounds produced by equipment that are not designedfor sound purposes. We could preliminarly suggest thatthe discipline of product acoustics will open up a wholeseries of new standards, with clear industrial andproject applications.
CONCLUSIONProduct acoustics refers to a number of
present needs and problems which, however, do notcorrespond to an existing discipline or professional.Therefore, it is likely that a systematic approach in thefoundation of this discipline will result in new trends ineducational, research and technological activities. Thispaper is a first attempt to share the ideas and instigatediscussion among acousticians.
AKNOWLEDGEMENTS
The author profited from fruitful andstimulating discussions with R. Murray Schafer, creatorof the World Soundscape Project, to whom this paperis dedicated.
REFERENCES
1. Dandrel, Louis. The Voice of Things, in IndustrialDesign, Reflection of the Century, Ed. by Jocelynde Noblet, Flammarion/PCI, Paris, 1993.
2. Schaeffer, P. Traité des objets musicaux. Paris,Seuil, 1966.
3. Gordon, J W. Perception of attack transients inmusical tones, PhD Thesis, Stanford University,1984.
4. De Poli, Piccialli et Roads (Eds.) Representationsof Music Signals. Boston, MIT Press, 1991.
5. Schafer, R. M. The new soundscape. Vienna:Universal Edition, 1969.
6. Cogan, R., Escot, P. Sonic design: The nature ofsound and music. Englewood Cliffs, NJ: Prentice-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
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 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 improve the comfort of aircraft passengers by the modification of those psycho acoustic and vibration parameters that physically correspond to the dimensions distinguishing combined acoustic and vibration perceptions in aircraft.
EVALUATION Following a literature study an adequate methodological instrument for the evaluation of acoustic and vibration experiences in aircraft did not exist, different field and laboratory pretests with an expert group as well with naive test persons have been carried out.[1,2,3,4] The aim was to develop a context orientated semantic differential (SD) (Table 1) concerning jet- and propeller airplanes on the one hand and helicopter on the other.[5]
MAIN TESTS
An aircraft simulation test was carried out in a mock-up (laboratory equipment) with a sample of 117 subjects (37 female and 80 male, aged 19 to 61). The helicopter test series have been carried out with 25 subjects (13 female and 12 male) taking part at the real helicopter flights and 107 subjects (45 female and 62 male) at the helicopter simulation tests in the mock-up, 25 of them took part before in the real flight test.
Table 1. Semantic Differential for Airplanes and Helicopter
ITEMS OF THE SD Heli-
copter Airplane
bearable unbearable X X comfortable uncomfortable X X threatening harmless X X
shaking calm X X vibrating not vibrating X X
dangerous safe X - pleasant unpleasant X -
oppressing liberating X - well-sounding ugly-sounding X -
crumpled smooth X - rotating still X - strong weak X -
shrill dull X - palpable impalpable X -
pushy reserved X - muffled not muffled - X
acceptable unacceptable - X regular irregular - X
monotonous varied - X high-frequency low-frequency - X
loud quiet X X rough not rough X X tonal not tonal X X
unsteady steady X X sharp not sharp X X
20 15
RESULTS
With the aid of principal component analysis (PCA), it was attempted to extract from the SD data independent perceptual dimensions describing the combined acoustic and vibration perception in aircraft. [6,7,8,9]
Jets and Props The factor explaining one third of the variance was a comfort factor related to aircraft interior sound and vibration. Comfort seemed to be the counterpart of specific sound characteristics such as the perceived loudness or roughness as well as particular vibration attributes (e.g. "vibrating"). The second dimension was associated with time characteristics of the flight situations like "monotonous" and "regular" and further
Figure 1. PCA for all propeller and jet flight situations vibration qualities like "unsteady" and "shaking". The third factor was confined to the perception of tonality.
Helicopter
By PCA of 5 flight situations 3 factors were detected describing the combined acoustic and vibration perception. The most important dimension explaining one third of the variance was a comfort factor related to specific noise characteristics such as loudness or roughness, vibration, and particular attributes describing the sensations. The second dimension was associated to further vibration qualities like rotating and shaking. The third factor was confined to the perception of psycho acoustic parameters.
CONCLUSION The process conforming the evaluation procedure to the objectives led to two different semantic profiles concerning airplanes and helicopters. The major innovations were in: the combination of noise and vibration, the novelty of the semantic attributes, the
incorporation of traditional psycho acoustic parameters, the validation by a thorough design procedure, and the availability in three major languages: German, English, and Italian. Following the results of the psychological and psycho acoustical research acoustical comfort in airplanes is everything contributing to the well-being and it constitutes an improvement of given conditions in an airplane. Acoustical comfort contributes to a general comfort definition by parameters which usually describe comfort, but as counterparts to the comfort parameters special acoustical parameters play a significant role.
ACKNOWLEDGEMENT BRITE-EURAM project "IDEA PACI" / BE97-4056
REFERENCES
1. Janke, W. & Debus, G., Die Eigenschaftswörterliste:
EWL; eine mehrdimensionale Methode zur Beschreibung des Befindens. Hogrefe, Göttingen, 1981.
2. Osgood, C.E, Suci, G.J. & Tannenbaum, P.H., The
measurement of meaning. University Press of Illinois, Urbana, 1957.
3. Osgood, C.E., Focus on meaning. Mouton, The Hague,
1976. 4. Pineau, C., The psychological meaning of comfort.
International Review of Applied Psychology, 31, pp. 271-283, 1982.
5. Quehl, J., Schick, A., Mellert, V., Schulte-Fortkamp, B.,
Remmers, H., Effects of helicopter and aircraft interior noise and vibration an passengers’ comfort sensation and subjective well-being. J. Acoust. Soc. Am., 105 (2) (1084) and ACUSTICA/acta acustica, 85, p. 158 ( 1999).
6. Quehl, J., Schick, A., Mellert, V., Schulte-Fortkamp, B.,
Remmers, H., Evaluation of combined aircraft interior sound and vibration effects on passengers` well-being and comfort sensation: the elaboration of a concept-specific methodologoical instrument. Results of the 8th Oldenburg symposium on psychological acoustics. bis, Oldenburg, 2000.
7. Quehl, J., Schick, A., Mellert, V., Schulte-Fortkamp, B.,
Remmers, H., Hauptdimensionen einer kombinierten Geräusch- und Vibrationswahrnehmung in Flugsituationen: Auswertungen zum semantischen Differential. Fortschritte der Akustik - DAGA 2000, Oldenburg, 2000.
8. Remmers, H., Reckhardt, C. & Bellmann, M., A system
of natural reproduction of sound and vibration. J. Acoust. Soc. Am., Vol. 105 (2), Pt. 2. (1999)
9. Zwicker, E., Psychoakustik. Springer, Berlin, 1999.
unsteady
shaking
vibrating
threateningsharp
rough
loud
monotonous
regular
muffled
comfortable
bearableacceptable
-1
-0,8
-0,6
-0,4
-0,2
0
0,2
0,4
0,6
0,8
1
-1 -0,5 0 0,5 1
Factor 1 [37.3%]
Low Frequency Perception in Urban Soundscapes.A Cognitive Approach
CatherineGuastavinoa, DanièleDuboisb,Jean-DominiquePolacka andChristineArrasc
aLaboratoired’AcoustiqueMusicale, UniversitéParis 6, 11 ruedeLourmel,75015Paris, FrancebLaboratoireCognitionsPratiqueset Ergonomie, 44 ruedel’Amiral Mouchez,75014Paris, France
cAcouphen,BP2132,69603VilleurbanneCedex, france
An investigationof thesubjective impressionof low frequenciesdueto traffic noisein urbansoundscapeswascarriedout,usingopenquestionnairesin threedifferentcontexts to betterunderstandhow low frequenciesaffect peopleoutdoors.A preliminarysurvey wassentby mail. Peopleweretheninterviewed in actualoutdoorenvironments,which wererecordedsimultaneously. The recordingswereusedfor listening testsin an acousticallydampedroom. Presentedin this paperarethe main resultsthat canbe drawn fromthepsychologicalexplorationof cognitive categoriesrelatedto low frequency phenomenaandtheir representationsin language.Thecomparisonof theresultsobtainedin thedifferentcontexts sketchessometheoreticalandmethodologicalissues.
OVERVIEW
In urban areas,noise stemsfrom a wide variety ofsources,many of which containpredominantlylow fre-quencies.Thesefrequencieseasilypropagateover largedistances,are able to proceedaroundobstaclesand filltheurbanspacecompletely. Complex multisensorypro-cessesareinvolvedin theperceptionof thelow frequencyrange(below 200Hz) whereaudiblesoundsbecometac-tile vibrations. The increasingproblemof noiseannoy-ancerevealsthelimits of physicaldescriptionof noisetomeasurethe subjective impression,andsuggestsa morecognitiveapproachto noisesasmeaningfuleventsthataf-fectpeople.
People living in three French cities (Paris, Lyon,Nantes)were questionedabouttheir appraisalof urbansoundscapesandtheirdescriptionsin threedifferentcon-texts. A preliminarysurvey wassentbymail to 80people.Answersrefer to memorizedrepresentationsof familiarurbansoundscapes.A shorteroral versionwasusedfor42 interviews in real outdoorsenvironments,whereallsensesareinvolved. Recordingswerecarriedout simul-taneously. The soundscapeswereselectedfrom a list oflocationsidentifiedasrepresentativeof city noises(Paris)[1, 2]. The recordingswereusedfor listeningtestswithno visual information, in an acousticallydampedroom,usinga stereoset-up(StuderA1) anda subwoofer (JBL4645C) below 100Hz. 29subjectslistenedto 6 differenturbansoundscapes.
In all three cases,semi-structured(’semi-directifs’)questionnairesweredesignedwith the sameopenques-tionsusingvery generalterms(’feel, beaffected’) in or-dernot to influencethe judgmentor confinetheanswersin predefinedcategories.Thepresentstudyrelieson theanalysisof psycholinguisticprocessesthat mediatebe-
tweenindividualrepresentationsin language(in ourcase,linguistic devicesinvolved in the descriptionof low fre-quency phenomena)andsharedconceptualcognitiverep-resentations.
RESULTS
Presentedbelow arethemainresultsthatcanbedrawnfrom the psychologicalexploration of cognitive cate-goriesrelatedto low frequency phenomenaandtheir rep-resentationsin language.
The linguistic analysisconductedon our verbaldatashows that the low frequency phenomenaareperceivedthroughthe following two semanticcategories: “soundevents”thatcanbeattributedto anidentifiedsource,and“ambientnoiseof thecity” or “backgroundnoise”(’bruitde la ville, bruit de fond’ in French)whereno specificeventcouldbeisolated.
A largevarietyof linguistic deviceswereobservedinthe descriptionof ambientnoise. Subjectsfound it verydifficult to identify anddescribe.They describedit in avery globalmanner, asa wholeratherthansourcesemit-ting noise,andexpressedit in termsof effectsperceivedby the subjectbut also physical propertiesof the sig-nal. Thephysicaldescriptionsrefer to the timbre(’muf-fled, muted’), the temporalstructure(’continuous,per-manent’)andtheenvelopment(’it is all aroundme’). Re-gardingqualitative evaluation,theambientnoiseis cate-gorizedasnonnegative (’not unpleasant’)for mostpeo-ple, and even comforting for some. It is consideredasa sign of humanactivity, characteristicof city life andthereforewell accepted.
Soundeventsaredescribedin termsof sources(’en-gine,trucks’) andactionor movementof thesourcegen-
eratingthenoise(’ theengineidles’). Subjectsdescribedhow they wereaffectedby thesoundeventsusingdever-baladjectives,i.e. adjectivesbasedonaverb,whichrefermainly to a hedonicscale(’annoying’). Their judgmentof the acousticphenomenais closely linked to their ap-praisalof the sourceitself andwhat is semanticallyas-sociatedwith it. Few descriptionsof the soundevent isgivenin termsof physicalparameters(34%of theoccur-rencesreferto physicalpropertieswhereas67%describeeffectson thesubject1). A soundeventis somethingthatoccursand can be heardduring a particularinterval oftime,thusnotabstractablefrom timeandspace.Thetem-porality is implicitly delimitedin thesoundevent,whichmayexplainwhy nodescriptionof thetemporalstructureis givenby thesubjects.
Recentpsycholinguisticstudiesshow that acousticphenomenacaneitherbe processedasnoises,perceivedaseffectsof the world on the subject,or in a morean-alytic mannerassounds,perceivedasobjectsof the ob-jective world [3]. Resultsshow that thesoundeventsareclearly processedas noises,whereasthe ambientnoiseis moreabstractedfrom theobject-source(sincethepro-cessof sourceidentificationfails) and more frequentlydescribedasa soundby meansof its physicalproperties,but ontheotherhand,it is alsoprocessedasameaningfulphenomenonreferringto thepresenceof humanactivity.
Ecological validity
Some methodologicaldifficulties were encounteredwith thelisteningtests.Recordingof six differentsound-scapeswerepresented,each5 minutelong. Instructionswere given to direct the subjects’responsestrategy to-wardsan everydaylisteningsituation,so that they react,to someextent, asif they werein a actualsituation,ac-cordingto theconceptof ecologicalvalidity developedbyGibson[4]. For eachsoundscape,subjectswereaskedtospenda few minutesacclimatingthemselvesto therecre-ated acousticexperience,and then answeropen ques-tions.Therecordingswerereproducedona2.1format(2speakersanda subwoofer, 1.2 meteraway from the lis-teningspot)in anacousticallydampedroom. Theuseofthesubwooferwasonefactorof thetest,resultingin onlyhalf of the testexamplesusingit. A total of 29 subjectsparticipatedin theexperiment.
Most subjectswere impressedby the low frequencyrecreationevenwhenthesubwooferwasnot beingused.Theconclusionis thatthevisualsettingaffectedtheir im-pressionof low frequency. Subsequently, the listening
1 Oneanswermaygive riseto severaloccurrences
roomhasbeenredesignedsoasto removethevisualref-erenceof the loudspeakers,maskingthemfrom the testsubjectfor futureexperiments.
Thecomparisonof thedataobtainedin thethreecon-textsshowssimilarresultsasfarthesoundeventsarecon-cerned.Thisconfirmsthatthestereoset-upusedin previ-ousstudies[1, 2], alongwith theinstructionsgivento thesubjectsis ecologicallyvalid in termsof sourceidentifi-cation.However, regardingambientnoise,theresultsarequite different. Ambient noisewasmostly describedintermsof physicalparameters(59%of theoccurrencesvs.25%in situ) andnotin termsof effects(14%of theoccur-rencesvs. 56% in situ) andalwaysprocessedasa sound.Moreover, subjectscomplainedthatthey did not feelasifthey were"there" andthat the envelopmentprovidedbythe subwooferwasnot consistentwith the frontal imageof theotherspeakers.
These results show that the same acoustic phe-nomenoncould give rise to two different cognitive ob-jects, namelya noiseor a sound,that integrateproper-ties of mentalrepresentationsinto physicaldescriptionsof thestimuli. This theoreticalpoint shouldbetakenintoaccountin methodologyfor listeningtestssinceecolog-ical validity dependson the purposeof the study. Stepsweretakento overcomethelimitation of stereoreproduc-tion by usingmulti-channelreproductionin futureexper-iments to improve immersion,which turns out to con-tribute to the cognitive representationof ambientnoise.Neverthelessthe differencebetweenlistening testsandeverydaylisteningsituationsmayalsobe imputedto theartificial laboratoryconditionsandtherequiredprocessesof abstraction.
ACKNOWLEDGMENTS
This researchis financedby the FrenchMinistry ofEnvironment.
REFERENCES
1. Maffiolo, V., Caractérisationsémantiqueetacoustiquedela qualitésonoredel’environnementsonoreurbain. ThèseUniversitéduMaine,Le Mans(1999).
2. Vogel, C., Etude des signauxsonores d’avertissement.ThèseUniversitéParis6 (1999).
3. Dubois,D., Categoriesasactsof meaning: the caseofcategories in olfaction and audition, Cognitive ScienceQuaterly, 1, pp.35-68.(2000).
4. Gibson, J., The Ecological Approach to Visual Percep-tion, LawrenceErlbaumAsssociates,Hilldale, New Jer-sey (1979).
Detection threshold of a periodic phase shift in music soundR. 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 basedon the A-X-B paradigm showed that the human auditory system cannot detect musical sound change if the cycle period of phasechange is lower than about a few hundred Hz.
INTRODUCTION
The human auditory system has been believed to beinsensitive to sound phase. For example, it is difficultfor 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 complextones clearly depends on phase difference between com-ponents [2]. Accordingly, belief that the human auditorysystem is entirely insensitive to sound phase is inaccu-rate. Hence, insensitivity of human auditory system wasmeasured in terms of ability to detect periodical phaserotation in musical sounds. Phase rotation was artificiallygenerated by means of all-path filters.
PHASE MODULATION BY ALL-PATHFILTER
An all-path filter is expressed in the s-plane as follows:
H�s ���
s2 � ω0
Qs � ω2
0
s2 � ω0
Qs � ω2
0
� (1)
where Q and ω0 are parameters determining filter phasecharacteristics. Figure 1 depicts the phase characteristicsof all-path filters with different ω0’s. Phase shift of � π isrealized at frequencies corresponding to ω0. By changingQ or ω0 as a sinusoidal function, filter characteristics varyperiodically. 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 attwo frequencies.
EXPERIMENTS
The minimum frequency to detect phase rotation wasmeasured through a listening experiment. The A-X-Bparadigm was employed to measure ability to distinguishphase modulated sound from the original one. X always
0 0.2 0.4 0.6 0.8 1−2
−1.5
−1
−0.5
0
Normalized Frequency
Pha
se M
odul
atio
n (×
π ra
d)FIGURE 1. Phase characteristics of all-path filters with differ-ent ω0s.
0 200 400 600 800 1000−2
−1.5
−1
−0.5
0
Sample
Pha
se M
odul
atio
n (×
π ra
d)
Norm. Freq.= 0.1Norm. Freq. = 0.6
FIGURE 2. Phase rotation at two frequency components.
represented phase modulated sound. Either A or B is thesame 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, asong by a female singer, pink noise, and a pulse trainof 1ms width with 2ms cycle. Parameter Q in the Eq.(1) was set to one and two. Parameter ω0 was changedperiodically within a range of 8 kHz to 20 kHz. Soundpressure level of each stimulus differed slightly depend-ing on the sound type, from 74 dB(SPL) to 77 dB(SPL).
Five subjects, four males and one female, took part in theexperiments. All were in their 20’s with normal hearingacuity.
RESULTS AND DISCUSSION
Figure 3 shows results of experiments. Ordinates ofthese figures exhibit the rate of correct responses. The au-thors regard the rotation frequency at which 75% correctjudgments are realized as the detection threshold. Thedetection threshold for instrumental music was about 300Hz. It was about 100 to 250 Hz for the song by a femalesinger, which is lower than that for instrumental music.Pink noise exhibited the highest detection threshold ofabout 4 kHz under Q � 1 condition among all stimuli ex-amined. The pulse train, on the other hand, exhibited thelowest detection threshold.
Detection thresholds under Q � 2 condition were gen-erally higher than those under Q � 1 condition. This maybe due to the small frequency region where phase shiftoccurs becoming broader along with increased Q value.This would also result in decreased cues for detectionof phase modulation, particularly in a lower frequencyrange.
CONCLUSION
Ability of the human auditory system to detect any de-terioration of sound introduced by artificial phase rota-tion in high frequency range was measured. Results oflistening tests revealed that human beings cannot noticerotation frequencies under a few hundred Hz. From thetechnological point of view, this knowledge has potentialfor exploitation in developing an innovative watermark-ing algorithm for musical sound.
ACKNOWLEDGMENTS
This study was partially supported by a Grant-in-Aidfor Development of Innovative Technologies (Millen-nium Project, 12107) by the Ministry of Education, Cul-ture, Sports, Science and Technology.
REFERENCES
1. K. Ozawa, Y. Suzuki, and T. Sone, Monaural phase effectson timbre of two-tone signals, J. Acoust. Soc. Am., 93(2),1007-1011 (1993).
2. B.C.J. Moore, An Introduction to the Psychology of Hear-ing, Academic Press Ltd., (1989).
101
102
103
0
25
50
75
100
[instrumental music ]
Rotation Frequency (Hz)
Cor
rect
ans
wer
s (%
)
Q=1Q=2
101
102
103
0
25
50
75
100
[song by a female singer ]
Rotation Frequency (Hz)
Cor
rect
ans
wer
s (%
)
Q=1Q=2
102
103
104
0
25
50
75
100
[pink noise ]
Rotation Frequency (Hz)
Cor
rect
ans
wer
s (%
)
Q=1Q=2
100
101
102
0
25
50
75
100
[pulse train ]
Rotation Frequency (Hz)
Cor
rect
ans
wer
s (%
)
Q=1Q=2
FIGURE 3. The rate of correct responses at each rotation fre-quency. Respective panels correspond to instrumental music, asong by a female singer, pink noise, and a pulse train.
An Annoyance Meter for Squeak-and-Rattle DiagnosticsD. Dufournet a, P. Susini b, M. Slama b, S. McAdams b
a 01dB-Stell, Limonest, Franceb IRCAM-CNRS, Paris, France
For sound quality requirements in industrial production, it no longer suffices to qualify acoustic sources in terms oflevel and frequency. Notions of robustness, safety and product quality are more and more related to auditorysensations, notably as concerns components of automobiles and domestic appliances. This paper presents a newtype of device that offers an innovative alternative to the classical sound level meter. Dedicated to the real-timecharacterization of sound quality, it allows the measurement of quality related to the perceptions of consumers andtakes into account the evaluation context. Based on a light and autonomous system, the device offers a palette ofpsychoacoustic criteria which can be dynamically integrated into a model of annoyance and/or comfort in order toadapt it to the phenomenon under study. An example of a psychoacoustic method is presented in the framework of anational research project on comfort in automobile interiors. The laboratory work based on panels of listenersallows the development of a model that is immediately integrable into the device. The access to a database ofdownloadable and sharable models is also discussed.
INTRODUCTION
The improvement of acoustic comfort in passengercar interiors is currently a major concern of carmanufacturers and suppliers because it corresponds toan increasing demand from customers. The past effort consisting in improving noisereduction with the help of absorption and maskingdevices has led today to an increase in driversensitivity to transient sounds, namely “Squeaks andRattles” (S&R). These signals, produced by cockpits,seats, and accessories, due to vibrations of the vehicle,are synonymous with low quality and have importanteconomic consequences. The correction of these problems is often realised byhuman experts who decide on their seriousness and thenecessary actions to be taken. This subjectivity hasseveral drawbacks: non-stability of the judgements, noon-line production control, no possibility to establishlong-term monitoring on subparts of the vehicle. It isthus necessary to establish adapted sound-qualitymetrics and to propose the methods andinstrumentation that can measure them in real time. An 18-month collaborative research project,“SQUAD”, financed by the French Ministry ofResearch and Technology, brought together Frenchautomobile and parts manufacturers,psychoacouscians, and noise engineers, to study thesemetrics and to propose a prototype system, based on alight hardware solution: a Squeak-meter. This paper presents the psychoacoustic methods andthe main features of the instrument. Extensions andproject perspectives are also discussed.
PSYCHOACOUSTIC EXPERIMENT
Psychoacoustic methods were employed to answerthe following questions:• What is the minimum acoustic level of an S&Rsignal that will allow it to be detected over the ambientdriving noise?• How can the annoyance sensation due to a detectableS&R signal be characterised and measured?It does not seem realistic to evaluate these answers forall driving situations and for each possible S&R signal.Preliminary work conducted in collaboration with carmanufacturers and suppliers consisted in identifyinglife situations of interest and the main S&R signalsthat often appear. Real recordings on the road (lifesituations), completed by test bank recordings usingcar and component shakers produced a database thatwas used for the psychoacoustics experiments. The experimental protocol consisted of choosing apanel of potential customers for listening tests in anaudiometric chamber, using several mixtures ofambiant noise for various life situations and S&Rsignals. Classical psychoacoustic methods and testswere defined and applied: adjustment and adaptivemethods (3 down-1 up) for a detection study andpsychophysical scaling methods for annoyanceassessment. Two series of experiments based onpotential customers and audio experts were conducted,focusing on detection thresholds and annoyanceratings for each panel.The following figures present some of the resultsobtained in detection and annoyance experiments.
FIGURE 1. Masked detection thresholds obtained forthe customer panel (dotted line) and audio experts(continuous line) for presentation of ten S&R signalsin a 130 km/h motorway ambient noise. Vertical linesindicate ±1 standard deviation of the mean. Theexperts are more sensitive by about 3 dB, but theircurve has the same form as that of the customer panel.
FIGURE 2. Comparison of annoyance ratings(expressed in absolute units) between experts(continuous line) and customers (dotted line) for tenS&R signals in a 130km/h motorway situation. Somedifferences appear for some S&R signals: customersare more annoyed by some sounds than are the experts.
The results obtained from these psychoacousticexperiments were analysed using both principalcomponents analysis and analysis of variance toestablish relations with physical features of theacoustic signals. Two algorithms, based on an estimation of specificloudness (ISO 532 B) and various psychoacousticcriteria were developed. Strong correlations withhuman judgements were found (r>0.8 for detection,r>0.85 for annoyance) and are proposed to beintegrated into the real-time prototype.
ANNOYANCE S&R METER
The integration of previous results in the SymphoniePC-based system was realised in dBQMark (QualityMark) software. Based on a real-time specific loudness(20 ms window) and other metrics, this softwareproposes 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 movingmicrophones in front of components to be qualified;• real-time user metrics based on polynomial forms ofclassical criteria (Loudness, Sharpness, etc..)
FIGURE 3. dBQMark prototype software: real-timeproduction of quality note and S&R annoyancemetrics.
CONCLUSION
The SQUAD project has produced an operationalprototype that will now be validated in real situationsby car manufacturers and suppliers. With thisinstrument, an objective quality note can be producedfor each kind of S&R signal in several life situations.Applications include default tracking, long-termmonitoring, and systematic on-line characterisation ofcar interiors. Some extensions of the software will berealised with a multi-channel acquisition board inorder to establish automatic localisation of S&Rdefects.
ACKNOWLEDGMENTS
We thank Visteon, Renault and PSA for participatingin the SQUAD project and for bringing to bear theirknowledge of S&R treatment and analysis.
Measurement of the Noise Quality of Post-SortingMachines 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, ItalybAKUSTIK - 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 particularlytrue for high-technology devices, such as machines for the automatic sorting and addressing of mail items, in which case bothproducers and users accurately account for ergonomic aspects. So a research has been undertaken in order to construct a scaleof pleasantness of typical sounds generated in a wide variety of post-sorting machines, in different environments (testing roomand plant) and under different operating conditions. After a complex measurement campaign, a complete database of recordedsignals 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 informationfor a refinement of the testing procedures, in order to get a reliable and robust result.
METHODOLOGICAL PREMISE
Current industrial practice for assessing the noiseenvironment in working places is essentially based onthe measurement of weighted sound-pressure levels, as
required by the related technical norms [1]. Now thisquantity is known to be just a rough indicator ofactually perceived loudness [2], its use beinghistorically justified by its simplicity ofimplementation, especially by analog orientedinstrumentation [3]. At present several factors pushtoward a more elaborate approach, including theavailability of digital flexible and portableinstrumentation, the progress in psychoacoustic studies[2,4] and an increasing sensitivity towards ergonomicaspects of workplaces [5-7,11]. In view of ergonomicaspects, it seems sensible to look for some indicatoraccounting for the overall acoustic sensation, leadingto a pleasantness/annoyance judgement. At present, nouniversally accepted acoustic pleasantness index isavailable, so that a search for an appropriate oneshould be done for each class of sound objects ofpractical or of theoretical interest. In this way both aprogress in each specific application field and,
gradually, a more thorough understanding of thejudgement forming process may be pursued. To thatgoal three major steps may be foreseen [8-9]:1Identification of the class of sound-objects underinvestigation and selection of set of them able to
represent the whole class. 2 Assessing themeasurability of the pleasantness index over anappropriate (psychophysic) scale. 3 Defining areference (primary) measurement procedure. Ofcourse, due to the experimental nature of the workseveral interactions may be needed prior to obtainingreliable and robust results. The current state for thepresent research will be briefly presented in theparagraphs to follow.
POST SORTING MACHINES
In figure 1 is presented a general post sortingmachine configuration. After the manual or automaticinfeed (a) the input post is transformed in a train ofsingle envelopes in order to be able to read the address(b). Then the envelope runs over a delay line, until theaddress image is processed and decoded. Then it goesinto the proper output tray which, when full, can bedownloaded manually (c) by an operator, or
WRAP UP MACHINEROBOT FOR AUTOMATIC TRAYS DOWNLOAD
PACKETS DOWNLOAD
OPERATOR POSITION
FEEDERADDRESS READINGMANUAL SORTING TRAYS
MICROPHONE POSITION
Legend:
FIGURE 1. Schematic diagram of a post sorting machine
automatically by a robot (d). In this case post packetscan be wrapped out automatically (e) beforedownloading (f). The overall length of such a plantvaries according to the number of download elements,a common length is around 30m; such a plant can sortmore than 40000 letters per hour. Measurements havebeen performed both on operators’ positions and onadditional position which may be occasionallyreached, for instance for surveillance or maintenancereasons. In the research the noise patterns of arepresentative set of post sorting machines has beenconsidered.
ACOUSTIC DATA BASE
As mentioned in the premise, the measure we lookfor should be defined of a representative subset of theconsidered objects. So the first essential step is theconstruction of the database of reference acousticsignals. This phase has at least two essential points: -define the structure of the database; - actually acquirethe signal considering practical limitations involvingmachine accessibility. Measurement conditions areparticularly severe if one considers the dimensions andthe complexity of the involved machines. The structureof the data base has been defined according to machinetype, working condition and recording position. Atpresent it includes more than 200 signals.
TEST DESIGN
From the database a selection of a first tentativereference set have been made. It includes n = 40signals lasting T = 15 s each. Two kinds of tests havebeen implemented: (forced choice) pair comparisons,and magnitude estimation with an anchor signal. Aproblem at this stage was the high number of trialsrequired. It was faced in two ways: by clustering thesignals into 9 clusters each including signals that aftera preliminary listening were considered “similar”.Both tests use a random sorting mechanism, ensuringthat each cluster was treated the same number of timesas the others. This approach allows each individual toperforms as much trails as he/she likes, without gettingtired, whilst the results may be easily cumulatedwithout biasing the distribution; clustering allows alsoa quicker convergence on the set of clusters (i. e. on apartition of the space of object under investigation).
EXPERIMENTAL RESULTS
The result of the magnitude estimation test isshown in figure 2. It consists of about 500 scores, froma jury of 30 people. The figure presents the mean scorefor each signal; bars indicate the mean dispersion. Dueto cluster organization, sorting is uniform over
clusters, it is not over the set of the signals, sinceclusters have not the same number of signals.
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 38 390
10
20
30
40
50
60
70
80
90
100
Magnitude estimation test
signal number
mea
n sc
ore
FIGURE 2. Signals’ mean score from magnitude test
Preliminary results form both kind of tests have beenprocessed and compared. They present a generalagreement, even if there are discordances for signalswith similar magnitude. Probably there is the effect ofsome inconsistency in subjective judgements (notremoved) and a biasing effect of the clusters. On thebasis of these first results the investigation willcontinue according to the following lines: reduction ofthe set of signals under test by elimination of signalsnot well characterized; new magnitude estimation test,with a modified anchor definition, by randomlyselecting single signals; cluster reorganization on thebase of the results of these first tests.
ACKNOWLEDGMENTS
The authors are grateful to ELSAG Bailey SpA forgiving the possibility to perform the experimentalwork. Special thanks to Dr. Leonardo Roncarolo forprecious advice and encouragement.
REFERENCES1. ISO 11200:1995. Acoustics – Noise emitted by machinery and
equipment – Guidelines for the use of basic standards for thedetermination of emission sound pressure levels at a workstation and at other specified positions.
2. Zwicker E and Fastl H Psycho-acoustics, Springer Verlag,1999
3. Yang S J, Ellison A J Machinery noise measurement, OxfordUniversity Press, 1985
4. Cook P R Music, cognition and computerised sound, MITPress, 1999
5. Meister D The history of human factors and ergonomics,Lawrence Erlbaum Ass, 1999
6. Salvendy Handbook of human factors and ergonomics WileyII, 1997
7. Bridger Introduction to ergonomics, Wiley, 19958. Roberts F S 1979 Measurement theory, Addison Wesley,
Reading MA9. Rossi G B, Crenna F, Belotti V Measurement of perceived
noise quality for the ergonomics of post-sorting machinesIMEKO TC 18 Congress, Sapporo, 2001
10. Purghé Methods of psycho-physics and uni-dimensionalscaling (in Italian), Bollati Boringhieri, 1999
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, FrancebVibratec, 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 beenrecorded with a dummy head located in the bus meanwhile the bus was driven in a typical way, including manyusual sound events (idle noise, acceleration noise, cruising noise, opening and closing of doors, switching on an offthe air conditioning system). This sequence was submitted to listeners who had to continuously evaluate the comfortof 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
Most studies dealing with noise comfort intransportation are related to steady state noises,though such noises are not so often present in a roadvehicle, as traffic conditions forces the driver to slowdown or accelerate quite frequently. Someresearchers have developed methods to continuouslyevaluate sound loudness of traffic noise, as heard byresidents : Kuwano and Namba [1] , who used acategorical method, Weber [2] who used acontinuous categorical method. Moreover, Susini andMcAdams [3] asked listeners to evaluate loudness ofsound recorded in cars, using a continuouscategorical method and a cross-matching one.The goal of this study was to adapt one of thesemethods to the evaluation of noise comfort, which isa multi-dimensional parameter (as compared toloudness). It was applied to noise in a city bus, inorder to identify the least comfortable events thatpassengers can heard during a typical journey.
MEASUREMENTS AND PROCEDURE
A dummy head was placed in a bus driven withoutany passenger on a journey representing usualconditions of a bus (engine idle, run up, constantspeed at 50 km/h, coast down, opening and closing ofdoors, etc…). The overall sequence duration was 162s. This sequence was submitted to listeners throughheadphones in a quiet room.The continuous categorical method, as defined byWeber [3], was used. A potentiometer was given tothe listener; it indicated five categories (“verycomfortable”, “comfortable”, “a little bituncomfortable”, “uncomfortable” and “veryuncomfortable”). While hearing the sound, the
listener had to move the potentiometer in a positionhe could freely choose along the whole range.The test was driven by Matlab software running on aPC computer. An audio card played the sound; a low-frequency generator provided a sine signal, theamplitude of which was adjusted by thepotentiometer; this signal was then digitalised byanother audio card. Finally, the envelope of thissignal was computed, which gave the continuousevaluation of noise comfort from the listener.48 people participated in the test; they were mostlystudents, did not mention any hearing impairmentproblem and were paid for their participation. First ofall, they were presented the whole sequence; thenthey made two successive evaluations of it, separatedby a short questionnaire about the test and the noise.
RESULTS
Individual Behaviours
First of all, we had to be sure that the task could beachievable by listeners. This can be evaluated fromtwo facts :- each listener gave two continuous evaluations of
the same sequence. The correlation coefficientbetween these two evaluations is always good(upper than …);
- the time-averaged value of each listener'sevaluations is similar to the overall evaluation ofthe sequence given by the listener in thequestionnaire separating the two evaluations.
Therefore, it seems that listeners succeeded in thetask, though about half of them (44 %) found it"rather difficult" or "difficult".When looking at listeners results, it appears that twoindividual behaviour can be found (figure 1).
FIGURE 1 : Example of individual answers
Some listeners try to closely follow their evaluationof sound comfort with the potentiometer (bottomcurve of figure 1), whereas other ones "integrate"their answer on some categories (top curve of figure1). Such a difference had already been pointed out byWeber [2].
Average Results
The average continuous evaluation is presented infigure 2.
FIGURE 2 : Average evaluation
From this curve, the least comfortable events clearlyappear :- both accelerations are the most prominent ones
(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], thatlisteners may have followed the change oftonality, due to the engine run-up, rather thantheir appreciation of comfort during theaccelerations, so that these results must becarefully accepted;
- the opening and closing of the doors (near 65and 80 s) are clearly uncomfortable;
- as is the starting on of the ventilation system(near 90 s).
Relation with loudness
Figure 3 shows the loudness of the sequence,computed according to the ISO 532B procedure, bythe MTS software Sound Quality.
20
30
40
50
60
70
0 40 80 120 160
sec.
SoneGD
FIGURE 3 : Loudness (ISO 532B) of the sequence
From a comparison between figures 2 and 3, itappears that noise comfort is closely linked toloudness, as it is often the case. But loudness can leadto an overestimation of some events (as doors closingat 75 s) or an underestimation of other ones (as thestarting on of ventilation system at 90 s).
CONCLUSION
The method of continuous evaluation, which hasalready been proven to be useful for loudness studies,can also be used in studies dealing with aestheticaspects of sounds. It allows to identify, in a complexsequence, the events where comfort is reduced and,therefore, to indicate to engineers where to put effortson in order to improve noise comfort. This study isnow going on by focusing on such events.
ACKOWLEDGEMENTS
The authors are grateful to P. Deprez and M. Salson,from Ratp, who authorised the publication of thispaper.
REFERENCES1. Kuwano S. and Namba S., Psychol. Res. 47, 27-37
(1985).2. Weber R., "The continuous loudness judgement of
temporally variable sounds with an "analog" categoryprocedure", in Proc. Of 5th Oldenburg Symp. OnPsycholog. Ac., 1991, pp. 267-294.
3. Susini P. and McAdams S., Acta Acustica, 86, 515-525 (2000).
Product Sound as an Important Part of Product DesignC. L. Fog
DELTA, Division Acoustics & Vibration, Building 356, Akademivej, DK-2800 Kgs. Lyngby, DenmarkPhone +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 soundsources. 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 whereand how Product Sound can be implemented in the design and product development process. A method using the combination ofmarketing research, psychoacoustics, and optimal mechanical design will be presented.
INTRODUCTIONThe quality of a product perceived by users and
other observers in the vicinity of the product dependson a number of product attributes such as appearance,response to user activities, function, noise/sound,weight, smell, taste/flavour, and tactile characteristics.We talk about the sound quality, the visual quality, thetactile quality, the quality of user interfaces, etc. [1].
Although the Product Sound should not be treatedas an isolated phenomenon, it makes sense in manycases to optimise this characteristic to improve theoverall perceived quality and thereby the user satis-faction regarding the product.
Product Sound can be considered as information,which is relevant in relation to sounds in e.g. the userinterfaces of the product/system, but Product Soundcan also be considered a part of the total experienceusing a product/system, which is a more marketing-oriented approach.
The overall objective in product development is toutilise future consumers’ attitudes, expectations, andpreferences so that the sound from a product becomes apositive attribute to the user instead of an annoyingproblem. As all hearing persons can perceive acousticquality and thus can be said to be experts, there is agreat need for good acoustic design and development.
SOUND IN DESIGNA good product design combines functionality, ap-
pearance, and quality in an optimal way and adds to thepleasure of using the product. By consciously workingwith the sound in the design an extra dimension of qua-lity can be added and even in some situations used di-rectly as an important competitive parameter in mar-keting. The starting point of this work may be an ana-lysis of the role of sound in the mental acts andoperations performed in interaction with the product.The object will be to examine the match between actand auditive feed-back. Is anything inappropriate andto what degree ? [2]
It should also be endeavoured to determine a targetsound – without a goal it is not possible to know inwhat direction the design work shall take. There are anumber of tools to be used in this connection:� Benchmarking
- To further develop/improve an existing productthe market can be examined and competitors’products purchased in order to get a comprehen-sive view of how far they have come and whatsounds are popular.
� Positioning- Is there a special sound picture forming a kind of
reference as to perception of quality, cf. doorslamming of certain cars?
- Is our product positioned appropriately in relationto this reference?
- Is the Product Sound free from elements that un-dermine the total perceived quality?
- Is the Product Sound acceptable in the environ-ment of the product?
More information can probably be obtained fromthe marketing department which has often made – incollaboration with external opinion research institutes– investigations of the users’ attitudes to and expecta-tions from certain product categories. Maybe there areeven product developers and engineers in the firmwhose knowledge can be used in connection with de-termination of the target sound.
THE PRODUCT SOUND WHEELWe have created a model for optimising Product
Sound Quality, see Figure 1 [3].The outer path in the Product Sound Wheel de-
scribes the fundamental process of optimising the Pro-duct Sound Quality. First, alternative sounds from aproduct, simulated sounds, or sounds from similar pro-ducts are presented to a test panel. The panel gives itsresponse either in answering forms prepared for stati-stical computations or directly, e.g. by setting sliders orpressing buttons. The same sounds are measured by
analysers, software, etc., and a number of metrics foreach sound is the result. The metrics may be any rele-vant traditional noise measure or may be more psycho-acoustically related as loudness, sharpness, fluctuation,strength, roughness, etc., or any combination of these.
By graphical or statistical methods the connectionsand correlations between the two kinds of measure-ments are sought, and usually it is possible to describethe preferred sound by objective metrics. By analysis ofthe physical characteristics of the sound-generating me-chanisms, the necessary design changes to obtain thedefined values of the metrics may be implemented.Tools for “sound tailoring or sound engineering”, soundediting, and simulation exist, and the lower inner path isoften an attractive shortcut to test different versions ofpossible sounds for further analysis or subjective tests.
A systematic approach in the design of low-noiseproducts is suggested in [4].
A properly designed product sound is an effectiveform of communication providing information aboutquality, function, and condition of the product, espe-cially as regards durable consumer goods.
Product Sound can be advantageous to apply to:- enhance the perceived quality and even “brand
sounding” – operational and signal sounds- support the user’s ego-intensity - life style and
personal image- strengthen the position of the product in the mar-
keting campaignFurther, manufacturers’ identity and image can be
supported by the Product Sound – identity in a narrowsense and image in a wider sense.
CONCLUSIONIn summation, whether a product sound is attractive
is not determined by the sound alone and its relation tofunction, but also by what the user is accustomed to,what the competitors’ products do, and not least im-portant: what the surroundings are willing to accept.
In the future we expect that listening tests will beused as a tool in product development to a higher de-gree than now, both for objective auditive measure-ments and for subjective measurements - affective testswhere preferences are asked for.
Furthermore we expect an increased use of sound inconnection with simulation and virtual reality (VR);when designers by means of a VR Centre demonstratenew products, the sound is also evaluated in addition tothe visual perception. The consequence of differentsurfaces, joints, and materials can both be seen andheard!
To a greater extent marketing and market researchon good Product Sound will take place via the Internet.Sound Bars, where it is possible to hear the productsound of various products before buying will pop up atcertain retailers’.
In a new 5-year project called “Human Sound Per-ception” DELTA will work intensively with researchon some of these aspects and especially with how tomake measurements with test persons as an efficienttool in the optimisation of perceived Product Sound.
REFERENCES1. J. Blauert & U. Jekosch, Sound Quality evaluation
– a multilayered problem, EEA-Tutorium, Ant-werpen, 1996.
2. J. Bernsen, Sound in Design, Danish Design Centerand DELTA, 1999.
3. C. L. Fog, Optimal Product Sound: Design andConstruction Guidelines for Developing Productswith Desirable Sound Characteristics and MinimalNoise, Report SPM 144 (in Danish), DELTA, 1998.
4. ISO 11688: Acoustics – Recommended practice forthe design of low-noise machinery and equipment.Part 1: Planning, Part 2: Introduction into physicsof low-noise design.
FIGURE 1. The Product Sound Wheel – a model for optimising Product Sound Quality.