VOICES IN JAPANESE ANIMATION: HOW PEOPLE PERCEIVE THE VOICES OF GOOD GUYS AND BAD GUYS. Mihoko Teshigawara

Transcription

1 1. INTRODUCTION VOICES IN JAPANESE ANIMATION: HOW PEOPLE PERCEIVE THE VOICES OF GOOD GUYS AND BAD GUYS Mihoko Teshigawaa Depatment oflinguistics Univesity ofvictoia, B.C., Canada Japanese anime, an animation medium which depicts the wold as inhabited by good and bad chaactes, is wildly popula in Japan and othe pats ofthe wold. The few scholaly studies that have consideed this medium ae still in the development stage (Lent, 2001). The pesent study examines the voices of chaactes in Japanese anime, focusing on the aticulatoy and peceptual chaacteistics ofthe voices ofmale and female heoes and villains. Vocal steeotyping plays an impotant ole in animation: voices need to eflect the physical attibutes and pesonality taits of chaactes and the vocal steeotypes that consumes, filmmakes, and voice actos shae. Pevious studies on vocal steeotypes (Zuckeman & Miyake, 1993) eveal that people infe simila pesonality taits fom voices. Howeve, few studies have investigated the acoustic coelates of pesonality in speech (Aonovitch, 1976; Zuckeman & MiYake, 1993), and few have investigated the auditoy coelates identified by phoneticians and the coespondence between auditoy coelates and laypesons' peceptions (Biemans, 1998; van Bezooijen, 1988). This study will attempt to fill the gap between these two types ofstudy, using speech samples fom Japanese anime. Pio to the pesent pape, the autho identified auditoy chaacteistics of the voices of 88 chaactes (44 heoes, 42 villains, and two suppoting oles) fom 20 animated catoons, using Lave's (1994, 2000) famewok fo voice quality desciption. The following summaizes the auditoy chaacteistics identified acoss categoies (fo moe details, see Teshigawaa, 2003, in pess): 1. Heoes' voices exhibited an absence ofphayngeal constiction and the pesence ofbeathy voice. 2. The majoity of villains' voices exhibited phayngeal constiction and hash voice caused by tense layngeal tension settings; howeve, phayngeal expansion accompanied by loweed laynx was obseved in a majoity of female and some male villains. In ode to investigate whethe the identified auditoy chaacteistics contibute to people's peceptions of good and bad chaactes, Japanese laypesons' peceptions of selected speech samples wee examined in an expeimental setting. Speech samples wee selected based on the auditoy analysis esults egading the peceived phayngeal states and phonatoy settings fom a sepaate study (Teshigawaa, 2003, in pess). In addition to pototypical heoes and villains, heoes that exlnoited phayngeal constiction and hash voice and villains without these chaacteistics wee included in ode to specifically examine the oles of these auditoy chaacteistics in vocal steeotyping. Paticipants wee asked to ate thei impessions of catoon chaactes using tait items in the following fou categoies: physical taits, pesonality taits, emotional states, and vocal chaacteistics. It was hypothesized that paticipants would attibute less favoable physical taits, pesonality taits, emotional states, and vocal chaacteistics to speakes who exhibited phayngeal constiction/expansion no matte which oles they played in the oiginal catoons. Statistical analyses wee pefomed in ode to examine the elationship between the auditoy chaacteistics ofthe voices and paticipants' tait atings ofthe speakes. 2. METHOD 2.1 Stimuli In light of the auditoy chaacteistics summaized above, the 88 chaacte voices wee divided into two goups, epesentative and nonepesentative: epesentative meaning that chaactes exhibited auditoy chaacteistics appopiate to thei ole, and nonepesentative meaning that chaactes exhibited auditoy chaacteistics opposite to o simply atypical of thei ole. Within these two goups, chaactes wee examined accoding to ole, gende and age (adult vs. child). Fo example, villains showing eithe phayngeal constiction o

2 150 Mihoko Teshigawaa expansion wee categoized into the epesentative villain goup, while those showing neithe tait fell into the nonepesentative villain goup. Thee wee 16 possible goups: heo o villain (2) X gende (2) X age (2) X epesentativeness (2). Howeve, since thee was only one child villain (male) in the copus, this classification system yielded only 13 goups. Two speakes wee chosen fo each of 12 goups, with the exception of the child male villain goup, which had only one speake. In addition, two suppoting oles (one male child and one female adult) exhibiting the auditoy chaacteistics of villains' voices, that is, phayngeal constiction and hash voice, wee added in ode to see whethe they would be ated similaly to heoes o villains. Theefoe, the voices of 27 speakes in total wee chosen as the basis fo expeimental stimuli. Noisefee speech samples of these 27 speakes had been stoed on a pesonal compute fo auditoy and acoustic analyses fo a sepaate study (Teshigawaa, 2003, in pess). They had been ecoded fom VHS tapes of the catoons. In ode to elicit listenes' esponses to the voices independent of vebal content, it was necessay to mask the contents ofthe speech samples. Ofthe five contentmasking techniques investigated by Schee, Feldstein, Bond, and Rosenthal (1985), andom splicing was chosen since it etains voice quality infomation (the focus of this study), while the othe techniques mask it. Fist, in ode to ceate stimuli epesentative of each speake, speech potions poduced with a voice quality setting deviating fom the speake's nomal setting wee emoved, with the exception of chaactes who wee consistently angy o shouting. Intensities wee standadized acoss speakes so that the maximum intensity was between 70 and 72 db. Following pevious eseach using the andomsplicing technique (Fiend & Faa, 1994; Schee, et ai., 1985; van Bezooijen & Boves, 1986), afte emoving pauses, the digitized speech samples wee divided into 250ms segments. The fist and last 3 ms of each segment wee linealy attenuated to' zeo amplitude in ode to avoid the intoduction of tansients (Fiend & Faa, 1994). In ode to ceate a 5s stimulus fo each speake, ms segments wee pepaed and eaanged so that segments could not occu in the same elative ode in the spliced stimulus as in the oiginal. In ode to countebalance the effects of odeing, two stimulus odes (A and B) wee used: in A, the 27 speakes wee andomly odeed disegading the speake goups, while B was the evese of A. Fo each speake, the speake numbe was announced followed by the 5s stimulus; afte 1 s of silence, the same segment was epeated, followed by 70 s of silence. This gave paticipants a total of 81 s to ate each speake, which, accoding to pevious studies (Kido & Kasuya, 2001; van Bezooijen, 1995), is consideed sufficient to ate the 21 tait items selected in this expeiment. Paticipants wee given a pactice session in which they ated an additional thee speakes befoe ating the 27 taget speakes. 2.2 Pocedues and Paticipants Twentyone tait items wee initially chosen to be used in the questionnaie fo the ating session. English tanslations ae given fo the items as follows: gende (female o male); age goup (010; 1118; 1935; 3660; ove 61); physical chaacteistics ("big," "goodlooking"); pesonality taits, 11 in total, of which thee wee chosen fo thei petinence to heoes of Japanese anime in paticula ("selfless," "loyal," "devoted") (Levi, 1998), thee wee thought to be univesal chaacteistics ofheoes ("bave," "intelligent," "stong"), and five epesented each of the five factos in the NEO Pesonality Inventoy ("sociable," "calm," "cuious," ~'conscientious," "sympathetic") (McCae & Costa, 1987); emotional states ("positive emotion"); and vocal chaacteistics ("highpitched," "loud," "elaxed," "pleasant," "attactive"). The 19 adjectives wee ated on 7point scales, fom 1 (not at all tue) to 7 (extemely tue). In the questionnaie, a1119 adjective items wee gouped accoding to the categoy and wee neve mixed disegading categoies. The emaining two multiplechoice items, gende and age wee consideed to be categoies on thei own, yielding six categoies all togethe. In ode to countebalance odeing effects of tait categoies and items within categoies, two odes of the six tait categoies (I) gende, age, physical chaacteistics, pesonality, emotion, vocal chaacteistics; (II) age, gende, vocal chaacteistics, emotion, pesonality, physical chaacteistics and, whee applicable, two item odes within tait categoies (a, b) wee pepaed. These wee systematically combined yielding fou questionnaie types (i.e., Ia, Ib, IIa, lib). Thitytwo paticipants (15 males, 17 females; aveage age 22.8 yeas old) wee ecuited fom Nagoya Univesity, Japan and the vicinity. In total, eight expeimental conditions wee yielded, combining the two stimulus odes (A and B) and the fou questionnaie types (la, Ib, IIa, lib). Fou paticipants wee assigned to each expeimental condition, with the exception of the two goups that used Questionnaie la, in which five paticipants listened to stimulus ode A and thee listened to B.

3 ' Voices ofjapanese Animation Expeimental sessions wee un in goups of up to seven in a soundpoof oom in the School of Lettes building at Nagoya Univesity. Using a CD playe, the expeimente played a CD containing instuctions ecoded by the autho, a pactice session and the 27 taget stimuli. The same instuctions wee given in the questionnaie booklet as well. Paticipants wee told that they would hea two 5s contentmasked catoon speech excepts fo each speake, and they wee asked to ate impessions of the speakes' taits on 7point scales and choose appopiate goups fo gende and age. Afte the expeiment, paticipants completed a questionnaie including demogaphic infonnation about themselves and thei exposue to anime. Each session lasted less than one hou. 3. RESULTS AND DISCUSSION 3.1 Reliability In ode to examine the consistency of paticipants' tait atings, Conbach's alpha was calculated fo each of the 21 tait items acoss paticipants. SPSS vesion 11.5 was used fo the statistical analyses pefomed in this pape. As shown in Table 1, the Conbach's alphas wee vey high, anging between.90 and.99 fo all but two items (.87 fo "sociable" and.80 fo "positive emotion"). Intaclass coelation, which measues the eliability between any two pais of judges (paticipants in the case of the pesent study) as opposed to the aggegate eliability of all the judges measued by Conbach's alpha (Rosenthal & Rosnow, as cited in Hecht & LaFance, 1995), was also calculated fo each item acoss paticipants. As was the case with Hecht and LaFance (1995), the intaclass eliabilities wee lowe ( =.11 to.82) than Conbach's alphas, which indicates that thee was consideably moe vaiability in the judgments of tait items at the individual level Table 1 Reliability ofratings Tait item Conbach's alpha Intaclass coelation Gende Age Physical chaacteistics Big Goodlooking Pesonality taits Bave Selfless Loyal Devoted Intelligent Stong Sociable Calm Cuious Conscientious Sympathetic Emotional state Positive Vocal chaacteistics Highpitched Loud Relaxed Pleasant Attactive Note. Figues ae based on all 32 paticipants In addition, since an anomaly was found in a close inspection of the aw data, bivaiate coelations wee calculated fo each pai of paticipants and fo each paticipant elative to the aveage atings of all paticipants, sepaately fo each item. It was found that atings of one of the paticipants in the condition goup that had stimulus ode A and questionnaie Ia had significant coelations with the aveage atings fo only six out of 21 items (p

4 152 Mihoko Teshigawaa <.05). Accoding to the expeimente, this paticula paticipant complained of fatigue and boedom afte the expeimental session; theefoe, it was decided that esults fo this paticipant would be emoved fom the est of the analysis. Due to the asymmety of the stimuli in child villain goups, the following analyses ae confmed to the adult heo and villain goups, which have full contasts in gende and epesentativeness16 speakes, two fom each of the eight goups, i.e., heo o villain (2) X gende (2) X epesentativeness (2). (Suppoting oles ae not discussed in this pape due to space limitations.) In addition, among the 21 tait items, five items (gende, age, "big" fom physical taits, and "highpitched" and "loud" fom vocal chaacteistics) wee emoved fom the pesent analysis because it appeaed that factos othe than phayngeal states wee decisive in influencing paticipants' impessions ofthese factos. (See Teshigawaa, 2003, fo an analysis ofthe omitted speakes and items.) Table 2 Means ofselected 16 Items by Speake Goup Heo Villain Male Female Male Female Items Rep Nonep Rep Nonep Rep Nonep Rep Nonep Physical chaacteistic Goodlooking Pesonality taits Bave Selfless Loyal Devoted Intelligent Stong Sociable Calm Cuious Conscientious Sympathetic Emotional state Positive Vocal chaacteistics Relaxed Pleasant Attactive Note. Rep stands fo epesentative goups and Nonep fo nonepesentative goups. n = 2 fo each goup ~ ~ 3.3 Analyses ofvaiance: Heoes Vesus Villains In ode to examine whethe paticipants esponded to stimuli accoding to the auditoy chaacteistics of the voices, that is, in efeence to diffeences in phayngeal state, a seies of theefacto epeated measues ANOVAs with thee betweensubjects factos fo contol pwposes was caied out fo each item. The thee withinsubjects factos wee ole (heo o villain), gende (male o female), and epesentativeness (epesentative o nonepesentative). The thee betweensubjects factos wee the two stimulus odes and the two categoy and item odes in the questionnaie. Fo the pwpose oftype I eo potection, the Bonfeoni coection was used; theefoe, an alpha level.003 (i.e.,.05 divided by 16) was adopted instead of the standad.05 level. Table 3 summaizes the significant main and inteaction effects that emeged in the analyses. It also includes patial eta squaed (11 2) as an estimated effect size fo each main and inteaction effect. As can be seen in Table 3, the main effect ofole was significant fo 12 ofthe 16 items. Theefoe, it can be

5 ' ' ' Voices ofjapanese Animation said that the voices of heoes wee geneally peceived as having moe favoable physical and pesonality taits, emotional states, and vocal chaacteistics than those of villains. Main effects fo the factos of gende and epesentativeness did not emege in as many items as fo ole (six main effects fo gende, and thee fo epesentativeness). This esult seems easonable, given that it was not hypothesized that phayngeal states would diffeentiate the two gendes. In addition, epesentativeness conveys diffeent phayngeal states depending on ole; fo instance, nonepesentative chaactes include both heoes exhibiting layngeal sphincteing and villains exhibiting an open aiway. In othe wods, epesentativeness is meaningful only in elation to ole. A numbe of inteactions wee found between any combination of two of the thee factos and among the thee. Except fo "stong" and "cuious," the same pattens wee obseved in the diection of inteactions acoss items. In the inteaction between ole and gende, females wee ated significantly highe when they wee heoes than villains, which was the expected tendency; howeve, males' atings wee geneally consistent acoss oles o significantly highe fo villains than heoes (Figue I). Six items showed this tendency and they ae all pesonality tait items: "selfless," "loyal," "devoted," "intelligent," and "conscientious." (Fo "cuious," the diection of the inteaction was evesed; male heoes wee ated significantly highe than male villains, while female heoes wee ated significantly lowe than female villains.) This tendency was not pedicted; athe, it was expected that both male and female villains would eceive low scoes. Table 3 Results Fom Analyses ofvaiance ofpaticipants' Tait Ratingsfo Adult Heoes and Villains Tait items Pesonality Souce Good df looking Bave Selfless Loyal Devoted Intelligent Stong Role(HN) F 81.09** ** 32.93** 26.99** ] MSE Gende (G) 1 F ** ** 1] MSE Repesentativeness (R) 1 F 12.71* ** 1] MSE HN*G 1 F ** 19.12** 15.06* 29.87** ] MSE HN*R 1 F 87.58** * 22.28** ** 32.88** 1] MSE G*R 1 F 48.39** 12.98* 27.56** 19.11** 36.27** * MSE ] HN*G*R 1 F ** 16.60** ] MSE Othes 8 R*Q02 F 15.08* 1]

6

7 Q) tq C (\'$ Voices ofjapanese Animation 5.2,'l' & E (ij c:. '0>... Ci; E 0 Q) 1U. 4.4 (j) w 4.2, Gende Mae 4:0+1 Female Heo Villain Role ' ' (1) Figue 1. Inteaction between ole and gende fo the item "conscientious." :2 s:: m Q).. w E (ij... s:: '0, Co Q) E "C 0; w 3.2, 3.0 ' R'flpesentativenes s: Rept~sehtatjy~.'...' 2~8'+I N()epesentatNe _Heo Viltain ' Role Figue 2. Inteaction between ole and epesentativeness fo the item "calm." The inteaction between gende and epesentativeness emeged in ten items. In this patten, males wee ated significantly highe when they wee epesentative than nonepesentative, while females wee ated 'significantly lowe when they wee epesentative than nonepesentative (Figue 3). The items whee this patten emeged included "goodlooking," "selfless," "sociable,t' and "sympathetic." This patten was not pedicted. These two kinds ofinteaction involving gende appea to have been caused by the elatively high scoes ofepesentative male villains. Relative to aveage atings, epesentative male villains ae much close to heoes o nonepesentative villains than epesentative female villains ae to the othe goups except fo the item "stong." It would be inteesting to see whethe gende would play such a significant ole in a study using moe natualistic 155

8 156 Mihoko Teshigawaa samples as peceptual stimuli ,~; 4.8 (f.l c o; Q) E ""C Q) o.s en 4.2 UJ 4.0 Repese ntaiivemess. Repesentative Non.;.;epesentative Male Female Gende Figue 3. Inteaction between gende and epesentativeness fo the item "sympathetic." Theeway inteactions wee found fo thee items: "devoted," "intelligent," and "cuious." Of the thee, "devoted" and "cuious" shae the same diection of inteaction: male epesentative villains wee ated highe than nonepesentative villains fo these items, while fo females, the same inteaction diection was found as in the egula twoway inteaction between ole and epesentativeness. Fo the item "intelligent," the egula oleepesentativeness inteaction was seen fo males, wheeas fo females, a clea diffeence between epesentative and nonepesentative emeged fo villains, but not fo heoes. In addition, thee significant but unexpected inteactions between one of the thee factos and one o moe contol factos (i.e., stimulus ode, categoy ode, and item ode) emeged: an inteaction between epesentativeness and item ode fo "devoted" (p.001); an inteaction between ole and stimulus ode fo "calm" (p <.001); and a fouway inteaction among gende and all thee contol factos fo "elaxed" (p =.001). It is not easy to intepet these esults. Fo instance, the item "devoted" appeaed in the following two odes in the questionnaie: (a) loyal, devoted, intelligent; (b) cuious, devoted, intelligent. In (a), "devoted" appeaed fouth among all the pesonality taits, while in (b), it appeaed fifth. Accoding to the ANOVAesults, these two goups of paticipants ated epesentative chaactes moe o less similaly while they ated nonepesentative chaactes significantly diffeently, with the atings of the item ode (a) goup being highe than those of the goup using item ode (b). Intepeting the inteaction between ole and stimulus ode in "calm" is moe staightfowad. The stimulus ode A goup ated heoes and villains similaly, while the stimulus ode B goup ated heoes significantly highe than villains. It is appaent that the contol factos, possibly in combination. with othe factos that have not been assessed in the pesent analyses, may have played a ole in these inteactions. In ode to detemine whethe the contol conditions used in this study affect listenes' tait atings, it would be necessay to... eplicate the expeiment with a much lage numbe of paticipants in each condition goup. In addition, moe caeful contolling may be necessay in futue eseach. 4. CONCLUSION This pape attempted to coelate auditoy chaacteistics and laypesons' peceptions of the voices of heoes and villains in Japanese anime. Based on the auditoy chaacteistics of the voices of heoes and villains identified in Teshigawaa (2003, in pess), 27 chaactes wee selected as taget stimuli fom Japanese catoons. Thitytwo paticipants listened to contentmasked speech excepts of the 27 taget speakes and ated thei

9 , Voices ofjapanese Animation impessions of physical and pesonality taits, emotional states and vocal chaacteistics on 7point scales. A seies of theefacto epeated measues ANOVAs was caied out fo each of the selected 16 items in ode to examine whethe paticipants esponded to the stimuli accoding to the diffeences in phayngeal state among the chaactes., In addition to significant main effects fo the facto ole, a numbe of significant inteaction effects between any combination of two of the thee factos and among the thee also emeged. The inteaction between ole and epesentativeness emeged in the majoity of the inteactions obseved, which suggests that paticipants attibuted less favoable physical taits, pesonality taits, emotional states, and vocal chaacteistics to speakes who exhibited nonneutal phayngeal states (i.e., phayngeal constiction o moe than a slight degee of phayngeal expansion) egadless of the oles they played in the oiginal catoons. This patten eveals that the classification of auditoy chaacteistics into epesentative and nonepesentative based on the phayngeal states identified in Teshigawaa (2003, in pess) was valid. Main effects fo gende, and inteaction effects between ole and gende and between gende and epesentativeness also emeged fo a numbe of items, suggesting that gende also played an impotant ole in the peception of the chaactes. In ode to moe fully investigate the coespondence between auditoy coelates and laypesons' peceptions of the voices of heoes and villains in Japanese anime, it is necessay to examine statistical coelations between the two, an analysis which is undeway in Teshigawaa (2003). REFERENCES Aonovitch, C. D. (1976). The voice of pesonality: Steeotyped judgments and thei elation to voice quality and sex ofspeake. The Jounal ofsocial Psychology, 99, Biemans, M.,(1998). Poduction and peception of gendeed voice quality. Symposium 'Peceiving and Pefoming Gende,,Zentum fu intediszip/iniie Fauenfoschung, Kiel Gemany, 1998, Fiend, M., & Faa, M. J. (1994). A compaison ofcontentmasking pocedues fo obtaining judgments ofdiscete I affective states. Jounal ofthe Acoustical Society ofameica, 96, Hecht, M. A., & LaFance, M. (1995). How (fast) can I help you? Tone ofvoice and telephone opeato efficiency in inteactions. Jounal ofapplied Social Psychology, 25, Kido, H., & Kasuya, H. (2001). Tsujo hatsuwa no seishitsu ni kanenshita nichijo hyogengo: Chosyu hyoka ni yom chushutsu [Eveyday expessions associated with voice quality of nomal utteance: Extaction by peceptual evaluation]. Nippon Onkyo Gakkai shi [Jounal of the Acoustical Society of Japan], 57, Lave, J. (1994). Pinciples ofphonetics. Cambidge, UK: Cambidge Univesity Pess. Lave, J. (2000). Phonetic evaluation ofvoice quality. In R. D. Kent & M. J. Ball (Eds.), Voice quality measuement (pp. 3748). San Diego, CA: Singula. Lent, J. A. (Ed.) (200I). Animation in Asia and Pacific. Bloomington, IN: Indiana Univesity Pess. Levi, A. (1998). The new Ameican heo: Made in Japan. In M. L. Kittelson (Ed.), The soul ofpopula cultue: Looking at contempoay heoes, myths, andmonstes (pp. 6883). Chicago: Open Cout. McCae, R. R., & Costa, P. T., J., (1987). Validation of the fivefacto model ofpesonality acoss instuments and obseves. Jounal ofpesonality and Social Psychology, 52,8190. Schee, K. R., Feldstein, S., Bond, R. N., & Rosenthal, R. (1985). Vocal cues to deception: A compaative channel appoach. Jounal ofpsycholinguistic Reseach, 14, Teshigawaa, M. (2003). Voices in Japanese animation: A phonetic study ofvocal steeotypes of good guys and bad guys in Japanese cultue. Manuscipt ofdoctoal dissetation in pepaation, Univesity ofvictoia, Victoia, Bitish Columbia, Canada.

10 158 Mihoko Teshigawaa Teshigawaa, M. (in pess). Voices in Japanese animation. In Poceedings ofthe Westen Confeence on Linguistics 2002, Novembe 13,2002, Univesity ofbitish Columbia, Vancouve, Bitish Columbia, Canada. van Bezooijen, R. (1988). The elative impotance ofponunciation, posody, and voice quality fo the attibution of social status and pesonality chaacteistics. In R. van Hout & U. Knops (Eds.), Language attitudes in the Dutch language aea (pp ). Dodecht: Fois. van Bezooijen, R. (1995). Sociocultual aspects of pitch diffeences between Japanese and Dutch women. Language and Speech, 38, van Bezooijen, R., & Boves, L. (1986). The effects of lowpass filteing and andom splicing on the peception of speech. Jounal ofpsycholinguistic Reseach, 15, Yanney, A. D. (1993). Steeotypes and ecognition memoy fo faces and voices of good guys and bad guys. Applied Cognitive Psychology, 7, Zuckeman, M., & Miyake, K. (1993). The attactive voice: What makes it so? Jounal ofnonvebal Behavio, 17, ,