Acoustic measures of voice quality

Although voice quality is sometimes taken as synonymous with phonation, in its broader definition voice quality results from a combination of oral vocal tract settings, including settings of the lips, tongue body, and so on, as well as the laryngeal and glottal settings that comprise phonation (Esling Reference Esling and Chappele2012). The sweet voice style certainly involves particular configurations of multiple aspects of the vocal tract beyond the laryngeal source (based on videos of seiyuu performances, for example, this style consistently involves lip spreading, resulting in a ‘smiling’ quality). Nevertheless, what is perceptually and articulatorily most salient about sweet voice is its phonation.Footnote ⁶ Thus, the acoustic analysis of sweet voice given here focuses on quantitative analysis of phonation, following the practice of previous acoustically based studies of Japanese voice quality (Ito Reference Ito2003, Reference Ito2004; Ishi Reference Ishi2004; Kong et al. Reference Kong, Kiyoko and Beckman2014).

Phonation refers to the process by which the airstream is shaped by the configuration of the vocal folds and other parts of the larynx. Creaky phonation, for example, is produced through the adduction and thickening of the vocal folds, resulting in irregular, slow vibration (Laver Reference Laver1980). Although changes in laryngeal settings are reflected in the resulting acoustic signal, interpreting the complex movements of the vocal apparatus via acoustic data has proven surprisingly difficult; many different measures have been employed in the phonetics and speech pathology literature, yielding varying degrees of correlation with human perception (Maryn, Roy, de Bodt, van Cauwenberge, & Corthals Reference Maryn, Roy, de Bodt, van Cauwenberge and Corthals2009). These measures often suffer from some degree of ambiguity in discriminating among several possible articulatory sources of a particular acoustic pattern, such as breathiness versus nasalization (Ishi Reference Ishi2004). Another issue in the acoustic measurement of phonation is that the interpretation of differences in the acoustic signal are often reduced to a simplified framework in which common phonation types—creaky, modal, breathy—are laid out in a continuum based on the size of the aperture through which the airstream passes (Gordon & Ladefoged Reference Gordon and Ladefoged2001). In reality, laryngeal configuration involves more variables than simply vocal-fold aperture size.

In spite of these drawbacks, acoustic measurements can provide some indications of the physiological settings underlying this style. The most common measures used in studies of phonation are H1-H2, H1-A1, and H1-A3 (Stevens & Hanson Reference Stevens, Hanson, Fujimura and Hirano1995; Iseli, Shue, & Alwan Reference Iseli, Shue and Alwan2007; Bishop & Keating Reference Bishop and Keating2012; Keating Reference Keating2014). ‘H’ in this terminology refers to harmonic and ‘A’ refers to formant amplitude; numbers indicate where they occur in the signal (see Figure 1).

Figure 1.

Sample spectral slice indicating H1, H2, A1, A3, 2k, and 4k.

H1-H2, meaning the magnitude of the first harmonic relative to that of the second harmonic, correlates with the open quotient, the percentage of time during the glottal cycle that the glottis is open (Swerts & Veldhuis Reference Swerts and Veldhuis2001; Kreiman, Shue, Chen, Iseli, Gerratt, Neubauer, & Alwan Reference Kreiman, Shue, Chen, Iseli, Gerratt, Neubauer and Alwan2012). H1-A1 is a measure of the bandwidth of the first formant and H1-A3 correlates with spectral tilt, indicating the rate at which upper harmonics lose amplitude. For all three of these measures, higher values are consistent with breathiness, which is characterized by lowered tension in the vocal folds and a wider glottal aperture, resulting in a broader bandwidth for F1 and less energy in the upper harmonics (Gordon & Ladefoged Reference Gordon and Ladefoged2001). Higher values of these measures can also result from other vocal settings, however, such as falsetto (a type of phonation in which the vocal folds are stretched and vibrate at a higher frequency), which has proven quite difficult to distinguish acoustically from other phonation types (Podesva Reference Podesva2007; Keating Reference Keating2014).

As sweet voice is a professional vocal style, previous studies of professionally trained voices, including singers and radio broadcasters, are particularly relevant. Two characteristics of trained voices are most salient here: enhanced periodicity and enhanced energy in the upper harmonics. Periodicity refers to the regularity of the vibration produced by the vocal folds: low periodicity correlates with dysphonic voices and nonmodal phonation (Hillenbrand, Cleveland, & Erickson Reference Hillenbrand, Cleveland; and Erickson1994; Heman-Ackah, Heuer, Michael, Ostrowski, Horman, Baroody, Hillenbrand, & Sataloff Reference Heman-Ackah, Heuer, Michael, Ostrowski, Horman, Baroody, Hillenbrand and Sataloff2003). Trained singers are thought to achieve more regular periodicity by controlling the extrinsic muscles that maintain the larynx in a constant position (Sundberg & Askenfelt Reference Sundberg, Askenfelt, Bless and Abbs1983). Periodicity has been successfully measured using Cepstral Peak Prominence (CPP). CPP measures cepstral peak magnitude relative to the overall amplitude of the cepstral signal, calculated using a cepstrum created by Fourier transformation of a Fourier transformation spectrum (Garrett Reference Garrett2013). CPP correlates highly with perception of dysphonic voices, and is considered a more reliable measure of breathiness than other common measures such as relative amplitude of the first harmonic or spectral tilt (Hillenbrand et al. Reference Hillenbrand, Cleveland; and Erickson1994; Heman-Ackah et al. Reference Heman-Ackah, Heuer, Michael, Ostrowski, Horman, Baroody, Hillenbrand and Sataloff2003; Samlan & Story Reference Samlan and Story2011; Garret Reference Garrett2013; Brinca, Batista, Tavares, Goncalves, & Moreno Reference Brinca, Batista, Tavares, Goncalves and Moreno2014). Voices with higher CPP are evaluated more positively in several respects: most notably, higher values correlate with perceived sexual attractiveness in both genders (Balasubramanium, Bhat, Srivastava, & Eldose Reference Balasubramanium, Bhat, Srivastava and Eldose2012). Another measure related to periodicity is harmonic-to-noise ratio (HNR), also calculated using the cepstrum. HNR at various frequency ranges has been used as a measure of roughness, breathiness, and falsetto versus modal phonation (de Krom Reference de Krom1993; Shue, Chen, & Alwan Reference Shue, Chen and Alwan2010; Keating Reference Keating2014). Both CPP and HNR measures are employed in this analysis.

The second relevant feature of trained voices is upper harmonic energy: specifically, a phenomenon known as the Actor's Formant (AF). The AF is a spoken analogue of the Singer's Formant, both of which consist of a peak in spectral energy in the 3 kHz region (e.g. Leino Reference Leino1993; Sundberg Reference Sundberg1987, Reference Sundberg2001). The AF is thought to correlate with what is perceived as a ringing or resonant voice quality, perhaps resulting from a narrowing of the laryngeal opening (Master, de Biase, Brasilia, & Laukkanen Reference Master, de Biase, Brasilia and Laukkanen2008; Lin, Jayakody, & Looi Reference Lin, Jayakody; and Looi2009). In the realm of voice artists, Warhurst, McCabe, Yiu, Heard, & Madill (Reference Warhurst, McCabe, Yiu, Heard and Madill2013) found that male commercial radio broadcasters had a more prominent AF than public radio broadcasters or nonbroadcaster controls. The AF has not been consistently found in female voices, however (Master, de Biase, & Madureira Reference Master, de Biase and Madureira2012). AF is most commonly quantified by measuring the relative magnitude of the highest peak in the 3-4 kHz region with the highest peak lower in the spectrum; here, this measure is operationalized as 2k-4k, meaning the highest peak around 4 kHz subtracted from the highest peak around 2 kHz.

Inter-speaker analysis

The first analysis seeks to acoustically characterize the ten sweet voice performances. Perceptual categorization by the author of the phonation type of each token finds that most speakers employ a mix of modal and breathy phonation, with very few creaky tokens (Figure 2). As creaky tokens are so rare, subsequent analysis contrasts only modal and breathy tokens.

Figure 2.

Proportion of perceptually creaky, modal, and breathy tokens for ten sweet voice performances.

Only two speakers (SS & TY) use more than 50% breathy phonation. The breathiness in these roles appears to serve two rather different stylistic purposes, as Shimamoto is portraying a mother interacting with her young children while Touma is portraying an alluring young woman flirting with boys. The role of sweet voice in constructing character types is discussed below (Function of sweet voice in anime).

Mean values for each acoustic measure for all sweet voice performances are given in Table 2. In terms of the distinction between modal and breathy phonation, while not all speakers show the expected trend for each measure, overall these results are consistent with previous work: H1-H2, H1-A1, H1-A3, and 2k-4k tend to be higher for breathy phonation, while CPP and HNR are lower. Linear mixed-model regressions performed in Rbrul (Johnson Reference Johnson2009) identify phonation as a highly significant factor in predicting each measure, controlling for F0, time point, and speaker (see Appendix B). Although the H1-based measures fail to reflect expected trends for three speakers (NH, NJ, OS), the CPP and HNR measures are consistent across the board; thus, as in previous work, periodicity is found to be a more reliable correlate of perceived breathiness (e.g. Hillenbrand et al. Reference Hillenbrand, Cleveland; and Erickson1994).

Table 2.

Acoustic measures by phonation type, ten sweet voice performances.

The mean fundamental frequency (F0) of the ten sweet voice performances is briefly examined here.Footnote ⁸ As indicated in Table 2, the relationship between perceived phonation type and mean F0 is not particularly robust across speakers. Linear mixed-model analysis, however, identifies a small but significant positive correlation between breathiness and higher F0 when controlling for speaker (see Appendix B). This finding is somewhat surprising because low F0 is often associated with breathiness, a pattern that does not hold here (Honorof & Whalen Reference Honorof and Whalen2004).

Rather than only comparing the pitch of these sweet voice characters to the pitch of ordinary Japanese women, it seems most relevant to first interpret them in relation to the other voice types that commonly appear in anime. In Teshigawara's (Reference Teshigawara2003) study of the voices of anime heroes and villains, she finds two groups of pitch targets for female heroic characters: 320 Hz and 400 Hz. As shown in Figure 3, the pitch of sweet voice characters is generally lower than both of these targets. Removing the exceptionally high Touma Yumi and Satou Ai, who average at 307.92 Hz, the mean pitch of the sweet voice characters is 262.7 Hz. This finding is consistent with the intuition that, while sweet voice can occur at a range of pitches, it is most often distinctively low by female anime voice standards. As discussed in the following section on the function of sweet voice within the anime setting, sweet voice characters are not heroines but rather are used to portray older-sister types; this lower pitch target helps distinguish them from main characters and establishes their relative maturity.

Figure 3.

Boxplot indicating range of f0 (Hz) for ten sweet voice characters (labeled by seiyuu).

In comparison to the speech of ordinary Japanese women, sweet voice is relatively high: Masataka (Reference Masataka1992) finds that Japanese mothers have an average pitch of 177 Hz when speaking to adults. When addressing children, however, their pitch increases gradually from 175 Hz to 252 Hz as they produce more utterances trying to elicit a child response. By contrast, Ohara (Reference Ohara, Pasero and Braun1999) finds a much higher average pitch for adult women, with her population of five female university students using an average F0 of 261.6 Hz when addressing a professor and 253.2 Hz when addressing friends. Taken together, these studies situate the pitch of sweet voice in the higher end of the normal female range in Japanese. Nonetheless, in a media speech context, and particularly in anime, the sweet voice is relatively low.

Intra-speaker analysis

As indicated in Table 1, contrasting nonsweet styles were collected for four of the seiyuu studied. From a qualitative perceptual standpoint, the sweet and nonsweet styles were extremely different from each other, to the extent that it was difficult to detect that the same speaker was producing both of them. Examining changes within individual speakers' performance of sweet and nonsweet vocal styles may help address the question of what makes the sweet voice so distinctive, while controlling for possible physiological and stylistic differences between speakers.

In the perceptual analysis, both the nonsweet and sweet performances consist of a mix of modal and breathy tokens, with only one speaker producing a significantly different proportion of breathy and modal (Shimamoto Sumi, χ ²(1) = 30.404, p < .0001) (Figure 4). As was the case for the sweet voice performances examined above, the nonsweet performances contain only a handful of creaky tokens (from the same seiyuu, Noda Junko). The difference in styles cannot, therefore, be accounted for by a change in the proportion of phonation types.

Figure 4.

Proportion of perceptually creaky, modal, and breathy tokens for nonsweet vs. sweet performances.

Table 3 lists the mean acoustic measures calculated for the four seiyuu in their two different styles. Mixed-effects linear regression modeling was carried out in Rbrul for each of the acoustic measures among the four sweet and nonsweet performances, controlling for speaker, phonation type, F0, and time point (see Appendix B). Sweet versus nonsweet style was a significant factor in predicting all measures except for CPP. For H1-H2, H1-A1, and H1-A3, sweet voice style correlated with higher values. As shown in Table 3, this effect can be seen among both the modal and breathy tokens. In contrast, for 2k-4k, sweet voice correlates with lower values: this trend is consistent for all but one of the cells. This lower 2k-4k value indicates that the sweet voice style has greater energy in the upper harmonics: specifically, it provides evidence for the Actor's Formant. This is consistent with the perception that sweet voice is more resonant. Similarly, the HNR35 value is significantly higher for sweet voice than nonsweet voice. This measure is less consistent than 2k-4k, however, with two speakers showing higher HNR for sweet voice and two showing lower HNR. Finally, CPP shows no consistent difference between the two styles, with two speakers (SS and SN) displaying remarkably similar CPP values in both styles for each phonation type.

Table 3.

Acoustic measures by phonation type, four nonsweet vs. sweet voice performances.

The mixed model indicates a small but significant positive correlation between sweet voice and higher pitch (see Appendix B). As shown in Figure 5, however, it is evident that this is not a robust phenomenon, as only two of the speakers demonstrate this F0 pattern.

Figure 5.

Boxplots of mean F0 for four seiyuu nonsweet (gray) vs. sweet (white) vocal styles.

In sum, while not all performances showed the same trend, the intra-speaker acoustic analysis has found several significant differences between nonsweet and sweet voice performances: higher values for H1-H2, H1-A1, H1-A3, HNR35, and F0, and lower values for 2k-4k. These trends are not wholly consistent with a one-dimensional phonation scale explanation: for example, for increased breathiness we would expect higher values on the H1-related measures, but not lower 2k-4k or higher HNR. These results are also inconsistent with falsetto, which has a lower HNR than modal phonation (Keating Reference Keating2014).

In light of these acoustic findings, let us return to the overall perception of these nonsweet versus sweet performances. The sweet voice performances have lighter timbres; as noted earlier, to a layperson, they may be perceived as ‘higher’ voices although they are occurring at the same absolute F0. Most notably, the performances from the same seiyuu sound so different that they are easily mistaken for two different individuals. In order to achieve this shift, the seiyuu must be making dramatic changes in their vocal apparatus. Moreover, we can speculate that sweet voice involves an unusual vocal setting, as this style is not normally encountered in everyday speech and only appears among trained professionals.

The most obvious example of a phonation type less commonly used in speech is falsetto; in the production of falsetto voice, the vocal folds are stretched so that they become thinner and vibrate more quickly, giving the voice a lighter timbre. A falsetto explanation for sweet voice is appealing in some ways, but it is not consistent with the higher HNR or lower 2k-4k observed in sweet voice—more importantly, sweet voice simply does not sound like falsetto voice. Sweet voice has a more tense sound, with a resonance that is reminiscent of a trained female singer in the middle of her range. In fact, there is a term used in the study of singing voices for a register lower than falsetto: ‘head voice’ or ‘head register’. In head register, the vocal ligaments are elongated as in falsetto voice, but the vocalis muscle is tensed; the resultant sound is more resonant than falsetto (Hirano, Vennard, & Ohala Reference Hirano, Vennard and Ohala1970; Miller Reference Miller2000). Professional singers generally practice developing more resonance in their head register as a way to avoid a perceptible gap between modal and falsetto as their pitch rises (Cryar Reference Cryar2014). Thus, resonance in head voice is a crucial feature that distinguishes trained and untrained voices. Björkner (Reference Björkner2006) observed a higher open quotient in head register compared to chest (modal) voice in female singers; this finding is consistent with the higher H1-H2 acoustic measure found in the present analysis.

The category of head register is not generally used in the study of spoken phonation, where it is apparently grouped in with falsetto. Certainly, speaking in head register is not a common phenomenon. Nonetheless, the notion that one can be trained to speak in head register, as distinct from falsetto and chest register, is widely accepted in the vocal coaching field (Jahn Reference Jahn2011).Footnote ⁹ The fact that producing a resonant head voice generally requires vocal training may explain the observation that sweet voice is quite difficult to replicate and is not used by ordinary Japanese women. Indeed, the head register theory can account for several of our observations about sweet voice. Acoustically, the higher H1-H2 and H1-A3 values for the sweet voice performances are consistent with head voice. The perception of the sweet voice style as lighter is also explained by head voice, in which the vocal folds are thinned and tensed, resulting in a lighter timbre. Previous research suggests that listeners use voice quality cues to estimate the location of an unfamiliar speaker's F0 within that speaker's pitch range: for example, pitches with higher H1-A3 values are rated as being lower in the speaker's range (Honorof & Whalen Reference Honorof and Whalen2004; Bishop & Keating Reference Bishop and Keating2012). This suggests that what listeners are hearing when they perceive that the sweet voice is ‘high’ is not in fact that speakers are producing a high F0, but rather that the F0 is perceived as being relatively low in the speaker's overall pitch range due to its higher spectral tilt.

Summary of acoustic study results

This acoustic examination of the sweet voice style sought to characterize the physiological basis of the sweet voice by examining the performances of ten seiyuu. Sweet voice performances consisted predominantly of a mix of perceptually modal and breathy tokens, corresponding to significant differences in H1-H2, H1-A1, H1-A3, CPP, HNR, and 2k-4k measures. Mean F0 of sweet voice performances was lower than that of female anime heroines but comparable to the higher end of ordinary Japanese female voices. An intra-speaker analysis of four seiyuu performing nonsweet and sweet vocal styles found higher values for sweet voice for the measures H1-H2, H1-A1, H1-A3, and HNR. Lower values were found for 2k-4k. Mean F0 was higher for sweet voice for two of the four seiyuu. The use of head register is proposed as an explanation for the acoustic and perceptual characteristics of sweet voice style.

Given the difficulty in studying voice quality through acoustic data alone, it may be profitable in the future to expand this analysis by studying the articulatory process of sweet voice production via ultrasound, EGG, or other similar methodologies.

Sweet voice characters

After developing a sense of the physical properties of sweet voice, it is time to move on to examining how the sweet voice functions within the anime setting. Again, this analysis primarily makes use of the ten sweet voice performances listed in Table 1, as well as a handful of additional performances from similar programs viewed in the course of this research.

Female characters with sweet voices appear in various genres of anime, including both male-oriented (shounen) and female-oriented (shoujo) programs, targeting multiple age ranges. In spite of this, certain commonalities exist among the characters who have sweet voices. Four general principles concerning sweet voice character traits can be outlined as follows.

1. i. Sweet voice characters are primarily (relatively) older women in positions of traditional female authority. The roles of the ten sweet voice characters in Table 1 include three mothers, two student council presidents, one older sister, two older girls, and one interior designer. Other sweet voice character occupations include teacher, school nurse, and goddess. The one nontraditional occupation in the ten performances is Carmen 99 (Gun X sword, 2005), a kind-hearted bounty hunter.

2. ii. Sweet voice characters are extraordinarily beautiful. Due to the stylized nature of the representation of the human form in anime, it is often difficult to determine by looking at a character whether they are meant to be attractive. Thus, voice quality provides a valuable cue to the audience in determining how a character's appearance is perceived by others. Characters with sweet voices are often beautiful to the degree that they elicit remarks of admiration, for example, “I've never seen such a beautiful creature before!” (Petopeto-san). Even mothers with sweet voice, who one might imagine are excluded from the heterosexual marketplace, are explicitly stated to be beautiful; in Tonari no Totoro, for example, the mother is asked by her young daughter if she will have hair as beautiful as hers one day. The use of beautiful here is crucially different from cute—sweet voice characters are not cute, but rather possess a mature, classical beauty.

3. iii. Sweet voice characters are neither villainesses nor heroines. As illustrated in principle i, sweet-voiced women are generally positive supporting characters who take care of the protagonist (either male or female) in some fashion. These characters rarely undertake independent action, instead reacting to events around them, most often with calm resignation.

4. iv. Sweet voice characters are sexualized yet ladylike. With the exception of the characters in young children's films, the sweet voice characters have their sexuality highlighted in some respect, though their behavior is generally far from lewd. These characters also tend to dress conservatively, wearing long skirts. Of the ten sweet voice characters in Table 1, three are explicitly or implicitly nonheterosexual.

Readers familiar with anime tropes will recognize that the character qualities described above are not representative of the average female anime character. A more typical anime girl is well-represented by one of the nonsweet voice examples from Table 1, Nausicaä from the Miyazaki film Kaze no tani no Nausicaä, a spunky, intelligent young princess who saves her community and brings peace to the world. The sweet voice characters, with their calm maturity, stand out in a medium dominated by frenetic youth and the aesthetic of cuteness.

Interaction of Japanese Women's Language and sweet voice

In addition to the distinctive behaviors and character traits associated with the sweet voice, the use of sweet voice in anime is accompanied by a set of linguistic features known as Japanese Women's Language (JWL). Contrary to what its name suggests, scholars agree that JWL is not commonly used today by most Japanese women, but rather only by a subset, and only in certain contexts (Okamoto & Shibamoto Smith Reference Okamoto, Smith, Okamoto and Smith2004:3). Nevertheless, JWL continues to be associated with women's speech in Japan, and carries pragmatic implications of supposedly traditional Japanese feminine features including softness and nonassertiveness (Inoue Reference Inoue2006:12).

A selection of previously studied JWL features that occur in the sweet voice data are listed in Table 4. The presence of each feature in the ten sweet voice performances and four nonsweet performances is listed on the right.

Table 4.

Selection of JWL features with number of observations in sweet and nonsweet performances.

While these data are limited by the brevity and content of the analyzed passages, and by the lack of a full-scale variationist analysis, some clear trends are nonetheless present. The most frequently observed JWL features in the sweet voice performances studied are sentence-final rising ne and wa (features such as interjection ara are more heavily bound to context and thus more rare). The fact that ne and atashi each occurred once in the nonsweet clips may indicate that they are less marked than other JWL features.

As described in Okamoto & Shibamoto Smith (Reference Okamoto and Smith2008), two discourses are relevant in the discussion of JWL: norms dictating that women should use ‘polite, gentle, and refined’ speech and norms regarding which linguistic features index these traits (Okamoto & Shibamoto Smith Reference Okamoto and Smith2008:88). Within this framework, the authors observe that various linguistic features may be identified as contributing to a feminine style: for example, while many linguistic norms for JWL relate to using features from standard Japanese, viewed as more refined than dialects, feminine styles can also be indexed using certain regional dialects. Specifically, the authors observe cases in which polite forms of the Kyoto dialect are used to construct a feminine style on television dramas (Okamoto & Shibamoto Smith Reference Okamoto and Smith2008:102). Indeed, in the sweet voice corpus there is one example of a speaker who uses the Kyoto dialect (Mai HiME, 2004). This finding underscores the point that the features listed above are simply a subset of possible grammatical features that may be used to construct JWL.

The use of JWL in anime is particularly significant in that it introduces these features to young viewers who do not necessarily hear JWL from speakers in their everyday lives. In her account of her own experiences with JWL, for example, Inoue (Reference Inoue2006:8) identifies particular television characters, including a teacher and an older-sister character, as the earliest speakers of JWL she ever encountered. Thus, JWL's function in anime is not merely interesting as a reflection of societal patterns, but may be crucial in the creation and perpetuation of second-order indices that connect linguistic features with particular feminine traits in Japanese society.

Given that sweet voice characters consistently use JWL, we might ask whether sweet voice carries any social significance independently of JWL. Is it simply a part of the women's language package? In fact, an asymmetry exists between the use of sweet voice and JWL in anime: while sweet voice does not occur without JWL, JWL can occur without sweet voice. Moreover, when JWL does occur without sweet voice, and in particular when certain marked features occur, such as sentence-final wa, the pragmatic function of the JWL is often quite different than it is in the case of the sweet voice characters—it implies a negative character evaluation. To put it another way, it is only when JWL co-occurs with sweet voice that it is portrayed positively.

Characters who use JWL but lack a sweet voice often fall into the category of the spoiled oujousama, or princess. The spoiled princess is a comedic villain, a girl with money to burn on expensive feminine clothing and a fancy education, who seeks to portray herself as proper and refined, but is inevitably revealed to have violent, unladylike tendencies under the surface.Footnote ¹⁰ Her trademark is her theatrical high-pitched evil laugh, accompanied by a stylized gesture of her hand poised to one side of her mouth. Through her expensive upbringing, the spoiled princess has acquired features of JWL; she flaunts her socioeconomic privilege by making use of extremely formal, feminine language at all times. While she can use grammatical features to put on a front of false femininity, she lacks a sweet voice quality—her voice is quite ordinary and modal. No matter how she disguises her personality, her unpleasant internal character prevents her from having a sweet voice.

The nonsweet voice of the spoiled-princess type reveals an ideology of linguistic authenticity in which speakers are able to manipulate their own use of grammatical features, but cannot control their voice quality. This is corroborated by the observation that characters are generally unable to turn sweet voice on and off. Within the setting of anime, sweet voice cannot be faked; one must be an authentically sweet person to have a sweet voice.Footnote ¹¹ This system has a level of irony to it, since viewers are aware that sweet voice is always, in some sense, a fake voice, performed by seiyuu who are able to produce a multitude of different vocal styles. The spoiled-princess type also reveals a negative view of JWL; in anime, sounding like a traditional Japanese woman is not necessarily desirable.

If we consider the broader linguistic context of anime, it is not particularly surprising that JWL features would become a resource for indexing negative character qualities. Mainstream anime celebrates youthful values and rejects, to some extent, traditional Japanese values and the world of adult responsibilities. More specifically, the aesthetic of cuteness that pervades anime, with its Western orientation and romanticization of the childlike, can be seen as a rejection of traditional Japanese female roles (Kinsella Reference Kinsella, Skov and Moeran1995:249). In a marketplace in which the greatest cultural capital is held by the modern and the young, JWL, as an index of traditional femininity, becomes the way that ‘other people’ talk. Even in the case of sweet voice characters, JWL still marks the other, although these other people are mothers and sisters who are portrayed positively. In the case of the spoiled princess, JWL indexes the inauthentic other, a rich girl who misguidedly attempts to buy her way into a refinement that can only be naturally possessed.

A primary function of sweet voice within the anime setting, then, is to provide authenticity to feminine gender performance. We might also argue that the sweet voice, as an index of physical beauty, fills a role that is played by women's language in more mainstream Japanese cultural traditions. Washi (Reference Washi, Okamoto and Smith2004:87) cites Kotaka Otsuma, an early twentieth-century advocate of women's higher education, in her discussion of the promotion of women's language as a means of enhancing physical appearance.

As this applies to anime, when a lady opens her mouth and speaks without a sweet voice, it is the fair trappings of grammatical features that can appear false.

Character-focused fan perception

Although the female characters in anime do not bear a close resemblance to live human beings, they nonetheless inspire feelings among certain fans.Footnote ¹³ As a result, much discussion has been devoted to what exactly makes different types of characters attractive.

One of the outcomes of this discourse is the notion of moe. Moe has a few meanings, but here we use it to refer to the erotic appeal of an anime character. Underlying the discourse of moe is the assumption that the emotional attraction that one feels toward an anime character is distinct from the appeal of a real human being, and therefore worthy of its own system of classification and discussion. On the website Heisei Democracy, a blog covering Japanese popular ‘two-dimensional’ culture (i.e. anime and manga), an author named Shingo has published several ambitious articles related to moe, one of them titled ‘A discourse on moe, Part I: Introduction and a moetic taxonomy’. Within his extensive moetic taxonomy, we find a description of a class of characters called the ‘Yamato Nadeshiko type’.

In this excerpt, Shingo refers specifically to voice as a defining feature of this character type, supporting our notion that voice quality is a significant contributor to the identity of these sweet voice characters. In fact, he brings up one of the seiyuu from Table 1, Inoue Kikuko (the most famous of the seiyuu included in this study), and declares her voice to be a sufficient condition for any character to qualify as a Yamato Nadeshiko type.

What exactly does he mean by Yamato Nadeshiko? Yamato Nadeshiko, a metaphor for the ideal Japanese woman literally referring to a species of flower (the fringed pink), is a potent term with deep historical roots that remains in use today (Bullock Reference Bullock1994; Sugihara & Katsurada Reference Sugihara and Katsurada1999). The notion of Yamato Nadeshiko had its heyday in twentieth-century imperialist Japan, in which, in contrast with previous eras, Japanese women were touted as the anchors of the home and ‘mothers of the nation’, rather than viewed simply as inferior to men (Skov & Moeran Reference Skov and Moeran1995a). The chronicler of Japan, Lafcadio Hearn, elegantly summarized the Yamato Nadeshiko ideal while describing the appeal of the Japanese woman to a Western audience:

Given that 100 years separate the above description and the anime involved in the present study, it is almost alarming how closely the sweet voice characters adhere to Hearn's notion of the character of the Japanese woman. While these remarkable women have never existed in the real Japan, in the Japan of the imagination they are alive and well.

Returning to Shingo's description of the Yamato Nadeshiko type, another significant point he makes is that this class of character is the most mature of the moe categories. Moe characters are normally extremely young girls, and the Yamato Nadeshiko is the single exception, made ‘more beautiful’ by her maturity. What qualities have enticed these fans of youthful cuteness to move to a higher age bracket? While Shingo's description avoids explicit discussion of sexuality, the appeal here is clear; regardless of how ladylike the Yamato Nadeshiko may appear, her maturity implies sexual experience. The sexual appeal of the older woman in Japan was noted by Clammer (Reference Clammer, Skov and Moeran1995:210) in his study of men's magazines, in which he observed a contrast between cute younger girls and the ‘true sensuality’ embodied by more mature women. This opposition of cuteness and mature beauty is one of the central distinctions that forms the structure of classification of sexually attractive women negotiated in the discourse of moe.

With the arrival of the Yamato Nadeshiko label, our sweet voice character, once a free-standing curiosity with an unusual voice, now finds herself interpolated into an elaborate structure in which she is contrasted with her younger, cuter peers, and connected to ancient female archetypes of the ideal woman. This system of female classification is in one sense unique to anime fan subculture, but in other ways part of a wide-reaching discourse implicated in numerous aspects of Japanese life. The phenomenon of categorizing the sexual appeal of women is hardly exclusive to anime fans, but moetic discourse is part of a distinctive subculture that engages in extensive quantification and standardization of the consumption of female characters, to the point of developing notions such as the zettai ryouiki, ‘absolute area’, meaning the distance of bare skin between the hem of the skirt and the top of the socks, for which the ideal ratio has been calculated to be 4:1:2.5 with a margin of error of ±25% (Dark Mirage 2008). This penchant for detailed classification is reminiscent of another Japanese subculture, that of advertising agencies, which in the 1980s yielded a plethora of classifications for types of female consumers, such as the ‘yenjoy gal’, the young working woman who likes to spend money during her leisure time (Skov & Moeran Reference Skov, Moeran, Skov and Moeran1995b:37). The identification of the Yamato Nadeshiko type, then, can be interpreted both in relation to the array of contrasting anime character types and to other subtypes of women who have been classified, measured, accounted for, and consumed in ancient and contemporary Japan.

Seiyuu-focused fan perception

A slightly different class of fan discussion surrounding the sweet voice focuses on the voice actress performing the vocal style, rather than on the character for whom the voice is performed. While not all successful seiyuu become big stars, some of them are able to parlay their success in performing particular roles into personal fame, moving from voice acting into recording music and making personal appearances. One seiyuu who has managed this feat particularly well is Inoue Kikuko, who has carved a successful niche for herself based on her sweet voice performances. Accounts of Inoue's voice written by fans commonly include both perceptual descriptors (e.g. ‘soft’, ‘sweet’) and descriptors which reference typical sweet voice roles (e.g. ‘motherly’, ‘angelic’).

Some fan descriptions of Inoue's sweet voice performances draw a connection between Inoue's natural speaking voice and her professional sweet voice. While fans are aware that Inoue performs both sweet and nonsweet voice roles, they consider the sweet voice performances to be closer to her true character.

It is no accident that fans draw this association between Inoue and the roles she performs; Inoue has deliberately created a public persona for herself that evokes the qualities of the sweet voice characters she has portrayed. As noted in the Seiyuu Directory quote above, Inoue has adopted the nickname onee-chan, a familiar term for older sister. She also, with much tongue-in-cheek fanfare, claims to be only seventeen years old, although she has been a professional seiyuu for longer than that (Manbow 2008). Like her characters, Inoue wears her hair long and dresses conservatively, preferring long skirts and long sleeves. The voice she uses for public events and interviews is not the sweet voice, but it is extremely breathy; we might call it a real-world approximation of the sweet voice. Although in some respects Inoue's public persona is reminiscent of the cute Japanese pop idols described in Kinsella (Reference Kinsella, Skov and Moeran1995) and Miller (Reference Miller, Okamoto and Smith2004), in the sense that she is creating an artificial character of a specific age and style in a calculated appeal to fans, the aesthetic of this ‘sweet voice idol’ is entirely different from that of a cute pop idol. Inoue is crafting a fantasy of a traditional, conservative Japanese young woman, one who wants nothing more than to settle down and take care of you. As shown in Shingo (2005), while this fantasy is not exactly cutting edge, it has its own distinctive appeal for fans.

For her fans, Inoue represents the Yamato Nadeshiko ideal, and central to that ideal is her voice:

Summary of fan perception

This examination of character-focused and seiyuu-focused fan discourse has demonstrated the prominent role of sweet voice in fan perceptions of Yamato Nadeshiko characters. These characters are older than the typically attractive female character type in anime, and their appeal lies in their sexual maturity and their embodiment of the ideal traditional Japanese woman who will care for you and yet be subservient to you. Fans of particular seiyuu show awareness both of the perceptual qualities of the seiyuu's voice and of the nature of the characters they portray. The example of the seiyuu Inoue Kikuko illustrates how voice actresses can exploit sweet voice performances to create successful idol personae marketing a fantasy of the perfect woman.