Predicing he perceived Qualiy of impulsive Vehicle sounds Marius Hoechseer, Jan-Michael Sauer BMW AG, D-80788 Muenchen, Germany. Ulrich Gabber Oo von Guericke Universiy Magdeburg, Universiaesplaz 2, D-39106 Magdeburg, Germany. Jesko L. Verhey Oo von Guericke Universiy Magdeburg, Leipziger Sr. 44, D-39120 Magdeburg, Germany. Summary A key elemen in he designing process of vehicle sounds is he evaluaion of he cusomers percepion and cogniive expecaion on he acousic qualiy. As he human audiory percepion comprises muliple dimensions, he degree of fulfilling he cusomers requiremens on acousic qualiy is a complex ask. In his respec, he percepion of impulsive sounds is paricular imporan since hose are he sounds he cusomers lisen o a heir firs conac when visiing a dealership. This sudy presens psychoacousic merics ha allow predicing heir perceived sound qualiy. In a firs sep, significan psychoacousic parameers were derived from verbal descripors assessing he sound qualiy of door handle snapback. Then, paired comparison ess and caegorical scale judgmens were carried ou for door closing and indicaor snapback sounds. Based on hese resuls, linear regression analysis revealed a significan predicive accuracy for wo psychoacousic parameers. The firs is he classical percenile loudness N 5 (see DIN 45 631) and he second is he newly considered measure duraion of sharpness. Their relaive conribuion depends on he considered sound ype. Whereas loudness appears o be he major predicor for he qualiy of indicaor snapback sounds, he duraion of sharpness has a large conribuion when predicing he acousic qualiy of door closing sounds. The influence of hese parameers was confirmed in addiional experimens measuring he qualiy of sounds where hese parameers were explicily varied. In summary, he daa show ha a linear combinaion of hese parameers wih signal ype specific adjusmens on heir significance can serve as a good basis for he qualiaive assessmen of vehicle sound qualiy. PACS no. 43.66.Lj, 43.66.Ba, 43.66.Cb, 43.50.Pn 1. Inroducion Since he echnical progress in auomoive developemen reached a high availabiliy of complex echnical funcions, he process of paradigm shif from raional o emoional aspecs has been acceleraed. In order o mee he cusomer s expecaions, he vehicle s acousic performance and appeal became decisive and imporan for he overall design process [1], [2]. In consequence, all perceivable qualiy crieria need o mach wih he general produc design [3], [4]. As sound qualiy is deermined on human percepion, he car manufacurers ake huge effors o evaluae he percepual characerisics of auomoive sounds wih respec o audiory sensaions [5]. An highly emoional experi- (c) European Acousics Associaion ence occurs he decisive momen of he cusomer s firs conac when visiing a dealership. In his silen surrounding, no disracing aspecs whils driving are presen and no engine is running. The relevan, mainly impulsive sounds resul from an explici acion of he cusomer and induce sensiive acion of he human audiory sysem. As hey include boh funcional and qualiaive informaion [6], [7], he inheren qualiy of his ype of sounds appears highly criical o he subjec s judgmen of qualiy. The aim of he presen sudy was o describe and predic he perceived qualiy of impulsive vehicle sounds using psychoacousic parameers. The parameers for he qualiy predicion were preseleced purposefully o correlae wih perceived sound qualiy according o facor loadings of verbal descripors [8]. As he imparial predicion required an inerval scaled classificaion, hearing ess were carried ou by he mehod of relaive comparison. These daa were furher vali- Copyrigh (2015) by EAA-NAG-ABAV, ISSN 2226-5147 All righs reserved 2411
0.2 objecive 0.1 subjecive ranking Facor a M. Höchseer e al.: Predicing he... EuroNoise 2015 luxurious loud solid sharp expensive hard b) Facor Fakor i i N S c) a) N X % II III S X % d) I IV 0 V 0.2 subjecive Figure 1. Schemaic skech of he course of he experimen. Recording of he simuli in a semi anechoic chamber and play back by open headphones a), verbal classificaion of qualiy b), deriving psychoacousical measures c), correlaion analysis of subjecive and imparial sound qualiy d). daed by an absolue ranking using a caegorical scale. Applying linear regression analysis, a significan predicive accuracy of wo psychoacousic parameers was revealed. The firs is he percenile loudness N 5 and he second is he newly considered measure duraion of sharpness, refered o T S in he following [9], [10]. The relaive imporance of he parameers depend on he signal ype: loudness appears o be he major predicor for he indicaor snapback sound whereas he duraion of sharpness seems o have an imporan role in he predicion of he sound qualiy of door closing sounds. As his approach included he semanic differenial o assess he subjecs demands on a high qualiy sound, psychoacousic parameers could be derived ha describe he audiory sensaion of he sound qualiy. Furhermore, a sandard operaing procedure o validae and imparially conrol he developmen process according o he cusomers ineress can be inegraed. 2. General Mehods 2.1. Recordings Door handle snapback and door closing sounds have been recorded in a semi- anechoic chamber wih he help of an arificial head (HEAD acousics GmbH, ype HMS IV) ouside he vehicle, as schemaically shown in Fig. 1a). The arificial head was placed wih he microphones 165 cm over ground and 45 cm aside and back o he car s b-pillar wih he face looking a he b-pillar. Regarding he door closing sounds, a closing velociy of abou 1.2 m/s (olerance: ±0.02 m/s) was chosen for he experimens. A ligh barrier was insalled in close proximiy o he vehicle o measure he closing velociy. For he recording of he indicaor snapback sound, a second aricifial head was placed inside he vehicle on he driver s sea. Considering he dimensions of each vehicle ype, an idenical posiion of he arificial head was assured wih he microphones beeing placed in fron of he car s b-pillar and above he seering wheel s upper edge. 2.2. Seup and procedure The aim of his sudy was o imparially predic he perceived sound qualiy of impulsive vehicle sounds. To reach his arge, he mehods of he semanic differenial, he paired comparison and he caegorical judgmen were used. Saring wih a semanic differenial es on he door handle snapback sounds, he semanic space was reduced o significan dimensions of percepion. As a crierion o choose, hese facors were used o preselec several psychoacousic parameers o predic he perceived sound qualiy of he door closing and he indicaor snapback sounds. Semanic differenial ess allow o assess he subjec s demands on a sound of high qualiy when measuring heir audiory percepion [8], [11]. This is schemaically described in Fig. 1b). A principal componen analysis was used o unie idenically associaed ad- 2412
EuroNoise 2015 Table I. Pairs of adjecives, choosen conex specifically for door handle snap back sounds according o welve expers of vehicle acousics. loud high frequency good sharp insable disurbing low value valuable inny uncomforable luxurious hard irriaing popping dull hollow cheap quie low frequency bad sof sable desirable high value worhless no inny comforable simple sof nice no popping meallic solid expensive jecives o overall dimensions of percepion. To derive significan facors, he scree plo s graphical urning poin combined wih eigenvalues of λ > 1 were applied as crieria as heir own variance exceeds he one of he conaining variable. On a saisical basis, Barle s es of sphericiy and he Kaiser-Meyer-Olkin crierion were applied. A facor roaion (varimax) compleed he analysis o raise he loadings. In advance o rank he perceived sound qualiy on an inerval scale (see Fig. 1d)), reliable and ordinal scaled daa resul applying he paradigm of paired comparison [12]. To ensure high concenraion during he evaluaion, he sounds were combined o each oher by he algorihm of Ross [13]. This seup included all simuli o appear equally ofen on he firs and he second posiion and o maximize he number of pairs beween he firs and he following playback of a sound. Furhermore, he individual preference in his form does no provide he acual difference in qualiy as he decision was made on he psychological coninuum. This has o be derived from he frequency, which liseners prefer one simulus o anoher [14]. Thus, he cumulaive resuls were ransformed o an inerval scale via dominance marix [14], [15]. Regarding he mehod of caegorical judgmen, a seven-poin Liker scale was used o assess he sound qualiy of door closing. Considering he liseners demands for he indicaor snap back sounds, he scaling was enlarged o a nine-poin Liker scale, also specified in equidisan seps. Based on he opposing adjecives low qualiy and high qualiy, he cenre of boh bipolar scales was defined as neural. As his scaling was prined on a shee of paper, he paricipans had o indicae he perceived sound qualiy. Each simulus had o be randomly evaluaed wo imes (es and rees) in order o check consisency of he answers. The sounds were presened o he paricipans over open head phones, ype STAX SR-202, as schemaically shown in Fig. 1a). Prior o an experimen, a wrien, sandardized ex was handed ou explaining he corresponding mehod and he es procedure. A he iniial phase of he experimen, he es manager presened he sounds o he paricipans. Aferwards, each quesion of he paricipans was answered by he es manager before he main experimen sared. 2.3. Paricipans In oal, sixeen women aged beween 23 and 40 years and 26 men from 19 o 52 years paricipaed in he experimens. The average age of hese liseners was 30.2 years wih all of hem reporing a normal hearing abiliy. Liseners had differen backgrounds: 28 were laymen and 14 expers working in he field of acousic developmen. Addiionally, he es design of relaive comparison considered he resricive crierion of conex independence whils making a choice [16]. I is valid as conex independen, if he enire scale values of he simuli are aken ino consideraion when making a judgmen. Based on he χ 2 -Tes, paricipans wih inconsisen responsiveness were excluded from furher analysis by a confidence level of α=0.05. 2.4. Psychoacousical parameers To predic he perceived qualiy of singular impulsive vehicle sounds, he presen sudy is focused on psychoacousical parameers. Considering he facor loadings of he semaic differenial es, he focus was on loudness (ANSI S3.4 [17], DIN 45631/A1 [18]) and sharpness (DIN 45692 [19], [20]). In paricular, he maximum gradien N max and average gradien N av of loudness N, percenile loudness N 5 (see Fig. 3) and he maximum sharpness S max were analyzed. An imporan derived measure is he duraion of sharpness T S. The duraion is calculaed as long as he signal exceeds a consan hreshold of sharpness level specified o 1acum [8], [9] (see Fig. 5). Due o he shor duraion and impulsiveness of he sounds, onaliy and roughness were no considered in he presen sudy according o he signals characerisics. 3. Experimens M. Höchseer e al.: Predicing he... 3.1. Experimen 1: Semanic space of door handle snapback sounds Simuli and Mehods For he semaic differernial es of door handle snapback, eigh sounds were recorded from differen vehicle ypes. Wih he help of welve acousic specialiss, conex specific adjecives were seleced including documened erms [21], [22]. Furher preess wih laymen resuled in a quesionnaire consising of he 17 pairs as seen in Tab. I. In order o collec valid resuls and o minimize habis, posiive and negaive associaed adjecives were irregularly disribued on boh sides of a bipolar, 2413
M. Höchseer e al.: Predicing he... EuroNoise 2015 Table II. Facor loading for Semanic Differenial es of door handle snap back. Facors: (1) Qualiy, (2) Inensiy, (3) Dynamic, (4) Timbre Facor Pairs 1 2 3 4 irriaing - nice -0,90 disurbung - desirable -0,84 cheap - expensive -0,81 low value - high value -0,81 valuable - worhless 0,77 uncomf. - comforable -0,77 luxurious - simple 0,73 loud - quie 0,88 hollow - solid -0,84 hard - sof 0,79 high freq. - low freq. 0,87 sharp - sof 0,82 insable - sable 0,67 good - bad 0,83 dull - meallic 0,80 inny - no inny -0,68 popping - no popping seven-poin scale. In addiion o ha, he sequence of he adjecive pairs was varied and he quesionnaires were changed depending on he simulus. Resuls In oal, he audiory sense included he four dimensions of percepion shown in Tab. II. The eigh sounds inveigaed covered a cumulaed variance of σ 2 = 79.07%. Based on he Kaiser-Meyer-Olkin crierion (0.77 < KMO < 0.84) and he Barle es of sphericiy (sig. = 0.00), he daa could be deemed as suiable and originae from a populaion only consising of correlaions o zero. The facor Qualiy shows he highes conribuion (σ 2 = 45.40%) o explain he percepual space. Based on a variance of σ 2 = 14.55%, he adjecives loud - quie, hollow - solid and hard - sof describe he second facor named Inensiy. The hrid one is called Dynamic as i includes adjecives relaed o frequency aspecs. The fourh dimension Timbre saisically offers an eigenvalue of λ = 1.39 and a variance of σ 2 = 8.19%. As he combinaion popping - no popping did no mee he crierion of normal disribuion, i could no be assigned o any facor. 3.2. Experimen 2: Sound qualiy of door closing sounds Simuli and Mehods The perceived sound qualiy of door closing was subjecively classified using he mehod of paired comparison. Therefore, foureen sounds were recorded and devided ino wo groups of eigh simuli. Two simuli were equal in each group. Wih he help of linear regression analysis, an overall psychoacousical predicion model based on loudness and sharpness was derived o correlae wih perceived sound qualiy. Figure 2. Correlaion analysis of subjecive and imparial sound qualiy of he door closing sound based on Eq. 1. Group 1 (lef), group 2 (righ). For furher validaion, addiional foureen sounds of door closing were evaluaed in an absolue manner on a seven-poin liker scale. Resuls The percenile loudness N 5 and he duraion of sharpness T S showed significan predicabiliy of he sound qualiy derived form he paired comparison daa (see Eq. 1). The euqaion (1) indicaes ha T S ( 0.21) is more imporan o describe he perceived qualiy of door closing sounds han N 5 ( 0.11). The correlaions for boh measures are negaive, i.e., a sof and moderaely sharp door closing sound is associaed wih high qualiy. SQ dcs = 0.28 0.11 N 5 0.21 T S, (1) Figure 2 shows he high correlaion resuls of he subjecive and he imparial sound qualiy r = 0.96 (group 1) and r = 0.95 (group 2). Raed by he mehods of he and he F es on a confidence level α = 0.05, he hypoheical assumpion is considered o be significan. For he psychoacousical measures loudness and duraion of sharpness, a hreshold crierion was used. Figure 3 shows how he overall correlaion of sound qualiy change wih his crierion, i.e., in his case wih he loudness percenile. For he door closing sound, he percenile value of N 5 had he highes correlaion o he perceived qualiy. Figure 5 shows he predicion accuracy for various sharpness levels which was he crierion for his measure. The hreshold of 1acum offered he highes values o correlae wih human percepion. Moreover, he derived model was used o predic he perceived qualiy of addiional foureen door closing sounds, classified absoluely via he caegorical judgmen. As he relaion shows a high coefficien of correlaion r = 0.94, his furher suppors a reliable and represenaive predicabiliy of his signal ype. 3.3. Experimen 3: Sound qualiy of indicaor snapback sounds Simuli and Mehods To classify he foureen indicaor snapback sounds, 2414
EuroNoise 2015 M. Höchseer e al.: Predicing he... Figure 4. Correlaion analysis of subjecive and imparial sound qualiy of he indicaor snapback sound based on Eq. 2. Group 1 (lef), group 2 (righ). Figure 3. Correlaion of loudness o sound qualiy over loudness percenile. Each symbol indicaes he correlaion of loudness (absolue value) o sound qualiy based on he corresponding percenile of loudness. The daa of he door closing and indicaor snapback sounds as well as heir arihmeic mean are shown. he same mehod of paired comparison was used for he door closing sounds. The daa were furher validaed by addiional experimens which quesioned he sound qualiy on a nine-poin Liker scale. Again, a correlaion analysis was used o evaluae he predicive qualiy of he significan imparial measures considering loudness and sharpness. Resuls The correlaion beween he inerval scaled resuls and he absolue ranking for he indicaor snapback sound was high (r = 0.94). Thus, he daa are regarded as valid. Based on he inerval scaling, linear regression analysis led o a psychoacousical predicion model wih he same measures N 5 and T S already used for he door closing sounds. For he presen signal ype, loudness is assigned wih a weighing value of 0.21 while he duraion of sharpness only had a weighing of 0.06 (see Eq. 2). SQ iss = 0.25 0.21 N 5 0.06 T S, (2) Using his model o predic he sound qualiy of indicaor snap back sounds (Fig. 4), he correlaion values of r = 0.98 and r = 0.97 concerning group 1 and 2 were even more significan han for he door closing sounds. One major aspec is he very high coefficien of correlaion for he percenile loudness N 5 as seen in Fig. 3. The duraion of sharpness is less imporan o predic he perceived qualiy, since he calculaed values of 0.65 and 0.66 in Fig. 4 correlae low wih he sound qualiy. Equal o he door closing sound, boh loudness and duraion of sharpness reach he highes correlaion a percenile N 5 and sharpness level of 1acum o describe he perceived qualiy of indicaor snapback sounds. Figure 5. Correlaion of he duraion of sharpness T S o sound qualiy over sharpness. Each symbol indicaes he correlaion of he duraion of sharpness (absolue value) o sound qualiy based on he corresponding hreshold of sharpness. The daa of he door closing and indicaor snapback sounds as well as heir arihmeic mean are shown. 4. Discussion As aesheic and emoional sound aspecs [2] conribue o he cusomer s whole picure of vehicle qualiy, he audiory percepion of he sound qualiy is affeced by muliple characerisics [1], [3], [5]. In order o imparially predic he relaed sound qualiy, relevan psychoacousical parameers need o be seleced purposefully. Considering he reduced dimensions of he percepion of impulsive vehicle sounds, he cusomer s preferred sensaion is refleced wihin four facors. Dependen on he share of variance, he facor Qualiy conribues o he subjec s expecaion on he qualiy of door handle snapback sounds, followed by Inensiy, Dynamic and Timbre. Thus, he audiory qualiy represens a complex measure, which is no angible by jus individual facors [8], [11], [20]. As he dimensions Inensiy and Dynamic include mainly adjecives relaed o loudness and frequency aspecs of sounds, he focus was on he psychoacousical measures of loudness and sharpness o predic he perceived qualiy of impulsive vehicle sounds. 2415
M. Höchseer e al.: Predicing he... The presen sudy showed ha for he imparial predicion of door closing and indicaor snapback sounds, wo major aspecs are sufficien. In agreemen wih published resuls on a mulidimensional semanic space [11], [23] and cogniive processing of impulsive vehicle sounds, a linear combinaion of wo psychoacousical parameers led o a higher predicabiliy of he perceived qualiy compared o a single measure [9] (see Fig. 3, Fig. 5).The same parameers for he percenile loudness N 5 and he duraion of sharpness T S were derived reflecing he similariy of he wo sound ypes. The relaive impac of he wo measures differed beween he signal ype. These resuls show ha he derivaion of a benchmark in sound desgin is a complex ask. Since he approach is direcly implemening he cusomers expecaions, muliple acousic aspecs have o be considered. 5. Summary and Conclusion The empirical sudies focused on he perceived qualiy of impulsive sounds when enering a vehicle. Based on hese models, sysemaically esable arges can be derived and inegraed in he developmen process of sound design. In fuure, specifically derived requiremens can be deermined for a cerain vehicle caegory o precisely saisfy he cusomer s ineress and o furher increase he appeal of vehicle comfor. References [1] H. van der Auweraer, K. Wyckaer, W. Hendricx: From sound qualiy o he engineering of soluions for NVH problems. Aca Acus. Acus. 83, 796-804 (1997). [2] U. Jekosch: Assigning Meaning o Sounds - Semioics in he Conex of Produc Sound Design. in: J. Blauer (ed.): Communicaion Acousics, 193âĂŞ219, (Springer, Berlin-Heidelberg-New York NY, 2005). [3] M. Haverkamp: Synäsheische Wahrnehmung und Geräuschdesign. in: K. Becker (ed.): Subjekive Fahreindruecke sichbar machen II, (Exper- Verlag, Renningen-Malmsheim, 2001). [4] J. Blauer: Produc sound design and assessmen: An enigmaic issue from he poin of view of engineering? Iner-Noise, Yokohama, Japan, 857-862 (1994). [5] K. Genui: Significance of psychoacousic aspecs for he evaluaion of vehicle exerior noise. J. Acous. Soc. Am. 123, 3134 (2003). [6] G. Bobber: Innengeräusche von Kraffahrzeugen. in: Forschrie der Akusik - DAGA 1988 (D. Ges. fuer Akusik, Braunschweig, 1988). [7] K. Genui, B. Schule-Forkamp, A. Fiebing: Neue Verfahren zum Benchmarking von Fahrzeuginnengeräuschen. in: K. Becker (ed.): Subjekive Fahreindruecke sichbar machen III, chap. 8, 127-145, (Exper-Verlag Renningen-Malmsheim, 2006). [8] M. Hoechseer, M. Wackerbauer, J. L. Verhey, U. Gabber: Psychoacousic predicion of singular impulsive sounds. ATZ - Auomobilechnische Zeischrif, (4/2015, Springer Verlag). EuroNoise 2015 [9] M. Hoechseer, M. Rolle, J. L. Verhey, U. Gabber: Physikalische und psychoakusische Vorhersage von singular impulshafen Geäuschen. 8. Symposium Moor- und Aggregaeakusik, Magdeburg, (2014). [10] M. Hoechseer, M. Rolle, J.-M. Sauer, B. A. Hemmrich, DE 102014212733.5. paen applicaion (7/2014). [11] M. E. Alinsoy, U. Jekosch: The semanic space of vehicle sounds: Developing a Semanic Differenial wih regard o cusomer percepion. J. Audio Eng. Soc. 60, 13-20 (2012). [12] N. Oo, S. Amman, C. Eaon, S. Lake: Guidelines for Jury Evaluaions of Auomoive sounds. Sound and Vib. 35, 24-47 (1999) [13] R. T. Ross: Scaling: A sourcebook for behavioral saisics. Aldine Pub. Company (1974). [14] F. Sixl: Messmehoden der Psychologie. Verlag Julius Belz, Weinheim (1967). [15] M. G. Kendall: Rank Correlaion Mehods. Griffin, London (1970). [16] R. D. Luce: The Choice Axiom afer Tweny Years. in: J. Mah. Psy., 15, 215-233 (1977). [17] ANSI S3.4-2007 (R2012) American Naional Sandard, Procedure for he Compuaion of Loudness of Seady Sounds (American Naional Sandards Insiue, New York, 2007). [18] DIN 45631/A1:2010-03 Calculaion of loudness level and loudness from he sound specrum - Zwicker mehod- Amendmen 1: Calculaion of he loudness of ime-varian sound; wih CD-ROM (Deusches Insiu fuer Normung e.v., Berlin, 2010). [19] DIN 45692:2009-08 Measuremen echnique for he simulaion of he audiory sensaion of sharpness (Deusches Insiu fuer Normung e.v., Berlin, 2009). [20] E. Zwicker, H. Fasl. Psychoacousics - Facs and Models. (Springer-Verlag Berlin Heidelberg, 3rd ed. (2006)). [21] E. Alinsoy: Knocking Sound as Qualiy Sign for Household Appliances and he Evaluaion of he Audio-Hapic Ineracion. in: Hapic and Audio Ineracion Design (Springer-Verlag Berlin Heidelberg, 121-130 (2012)). [22] T. Hashimoo: Die japanische Forschung zur Bewerung von Innengeräuschen im PKW. in: Zeischrif fuer Laermbekaempfung, 41, 69-71 (1994). [23] S. Kuwano, H. Fasl, S. Namba, S. Nakamura, H. Uchida: Qualiy of door sounds of passenger cars. Acous. Sci. & Tech. 27, 309-312 (2006). 2416