Vocal tract dimensional characteristics of

International Journal of Speech-Language Pathology, 2013; Early Online: 1–8

Vocal tract dimensional characteristics of professional male and female singers with different types of singing voices

NAN YAN1,2, MANWA L. NG2, MOK KA MAN2 & TSZ HIN TO2

Int J Speech Lang Pathol Downloaded from informahealthcare.com by University of Hong Kong on 02/17/13 For personal use only.

1Ambient

Intelligence and Multimodal Systems Lab, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, PR China and 2Speech Science Laboratory, Division of Speech and Hearing Sciences, University of Hong Kong, Hong Kong, PR China.

Abstract The present study examined the possible relationship between classification of professional singing voices and their vocal tract parameters including vocal tract length and volume, and vowel formant frequencies. Acoustic reflection technology (ART) was used to measure vocal tract length and volume of 107 professional singers: 32 tenors, 25 baritones, 27 sopranos, and 23 mezzo-sopranos. The first three formant frequencies (F1–F3) of the English vowels /a, æ, i/ produced by the professional singers were also obtained. Results indicated significantly shorter oral and vocal tract length, and smaller oral and vocal tract volume associated with sopranos when compared with mezzo-sopranos. Acoustically, sopranos had higher F1, F2, and F3 values than mezzo-sopranos. The present findings suggest that, in addition to vocal tract length, vocal tract volume may also affect formant frequencies, implying the possibility that classifying professional singing voices is based on both vocal tract length and volume information.

Keywords: Vocal tract, formant frequency, acoustic pharyngometer, professional singing, singer’s classification, soprano.

Introduction Traditionally, professional singing voices are categorized by experienced vocal pedagogues into at least three main singing types in each gender: bass, baritone, and tenor for male singers and alto, mezzosoprano, and soprano for female singers (Titze, 1994). Each singing voice possesses a unique set of attributes, and is based on which different professional singing voices are distinguished from each other, at least perceptually. Generally speaking, voice range and voice timbre are traditionally considered the main factors in determining a singing voice type by vocal pedagogues. Voice range refers to the range of the singing pitch that a singer can achieve (Titze, 1994). In general, a lower singing register such as bass is associated with a lower pitch range, whereas a higher singing register such as soprano often sings at a higher pitch range. Yet, defining different professional singing voices based merely on the pitch (voice) range may not be accurate as the voice range of different singing voices may overlap (Titze, 1994). For example, the voice ranges of baritones and basses overlap with each other in more than 78% of their voice range (from 98.0–329.0 Hz).

In addition to voice range, vocal pedagogues also classify professional singing voices based on passaggio, which is referred to register transition points. Most pedagogues agreed that differences in location of the passaggi reflect differences of structure and timbre between the difference singer voice types (Sundberg, 1996). Moreover, voice timbre seems to be of use to singer type classification. Voice timbre can generally be understood as when two tones are of the same pitch and loudness but are judged to be different, and then the two tones are said to have different timbres (American National Standards Institute, 1973). It follows that voice timbre is related to the perceived voice quality. Even if two singers exhibit exactly the same voice range, they can still be classified into two different singing voices if their voices are of different timbres. Erickson, Perry, and Handel (2001) suggested that timbre was a perceptual parameter used to distinguish between different singing voices, and “timbre transformation” was created by altering the singing voice register across a singer’s voice range. Sundberg (1987) defined a vocal register as a frequency range of phonation that consists of tones produced in a similar way, or voice

Correspondence: Nan Yan, PhD, 5/F Prince Philip Dental Hospital, 34 Hospital Road, Sai Ying Pun, Hong Kong. Tel: ⫹852-2857-1402. Email: nyan@ hku.hk ISSN 1754-9507 print/ISSN 1754-9515 online © 2013 The Speech Pathology Association of Australia Limited Published by Informa UK, Ltd. DOI: 10.3109/17549507.2012.744429


2

N.Yan et al.

timbre. Yet, voice timbre is a subjective way to classify different singing voices (Erickson et al., 2001; Erickson, 2003). Perception and judgement of voice timbre could be affected by a number of factors including listeners’ experience and skill levels. While researchers have been attempting to identify factors determining professional singing voices, some factors, such as the physiology of different vocal registers, and their corresponding frequency ranges, how these factors affect different professional singing voices, and how vocal pedagogues differentiate them are still not clear (Hollien, 1974; Miller, 2000). Agren and Sundberg (1976) compared the source spectra of five vowels (/a/, /e/, /i/, /ε/, /u/) sung by two altos and two tenors, and found no consistent difference in amplitude profile in the spectra between altos and tenors. Yet, they found significantly greater amplitude in altos’ first harmonic when compared with tenors’, based on which they suspected that amplitude (energy) of the first harmonic could be the main difference between the tenor and alto voices. Fundamental frequency appears to be a distinguishing factor for classification of different professional singing voices. Nonetheless, only four participants were included in Agren and Sundberg’s (1976) study, and comparison was made only between altos and tenors. Therefore, a convincing conclusion could not be drawn. Cleveland (1977) examined the five vowels (/a/, /i/, /e/, /o/, /u/) sung at four pitches (C3, F3, A3, E4) by eight professional singers perceived by vocal pedagogues. By analysing the pitch, formant frequency and source spectra of the singing samples, they reported a conclusion similar to Agren and Sundberg (1976): fundamental frequency tends to help differentiate various professional singing voices. Their results showed that, when classifying singers, pitch (fundamental frequency) served as the “primary cue” and formants as the “secondary cue”. Vocal pedagogues mainly depended on pitch when classifying the voice stimuli into a particular type of singing voice. However, when asked to classify voice stimuli of the same vowel and similar pitch sung by tenor singers, baritone singers, and bass singers, the vocal pedagogues had to rely on formants, as fundamental frequency became similar. Cleveland (1977) also found that the mean of the first four formant frequencies in tenors was higher than that of baritones, while bass singers had the lowest mean of the first four formant frequencies. As formant frequencies are largely determined by the vocal tract configurations, such findings suggest that vocal tract characteristics might be a contributing factor of professional singing voices. However, results of this study also lacked statistical significance as only eight participants were included. In another study, Berndtsson and Sundberg (1995) found that a singer’s formant was relevant to the classification of singing voices. The singer’s formant, also referred as the “singing formant” or “ring”, was an important acoustic feature observed in Western

opera singers (Vennard, 1967, pp. 144–145). Researchers later found that such singer’s formant is formed by the clustering of energy of F3, F4, and F5 (around 3 kHz). To date, the physiological mechanism of how the singer’s formant is produced is still under debate (Detweiler, 1994). Regardless, it is generally agreed upon that the singer’s formant is related to the configuration of vocal tract. For example, Berndtsson and Sundberg (1995) found that a higher singer’s formant was associated with the classification of singing voices of a higher pitch. Based on the above discussion, it is known that classification of singing voices depends on: (1) fundamental frequency, which is mainly affected by vocal fold vibration, and (2) formant frequencies, which are affected by the vocal tract. However, the relationship between formant frequencies and vocal tract and their interaction with singing voice classification is still uncertain. Dmitriev and Kiselev (1979) measured the vocal tract length of 20 Russian professional singers of different singing voices, namely bass, baritone, tenor, mezzo-soprano, soprano, and high soprano, with lateral x-ray, and they correlated the x-ray measurements with the corresponding acoustic characteristics (the high and low singing formants). It was found that each professional singing voice type had a specific range of lower formant frequency and higher formant frequency (see Table I), and their vocal tract length differed systematically according to the singing voice type; sopranos had a shorter vocal tract than mezzo-sopranos; tenors had the shortest vocal tract, followed by baritones, and basses (see Table I). This shows that professional opera singers only sing with a fixed range of vocal tract length for all vowels, even when the pitch and articulation are changing. There seems to be a pre-determined vocal tract length for each professional singing voice. However, previous studies focused only on how singing voices are correlated with vocal tract length. The radioactive and hazardous x-ray technology can only provide two-dimensional visual information of the human vocal tract. Volumetric data of the human vocal tract associated with different professional singers cannot be obtained. It is not known whether vocal tract volume also plays a role in determining voice timbre of different singing voices and classification of singing voices. Table I. Frequency of the low singing formant, frequency of the high singing formant, and the length of the vocal tract of different singing voices (Dmitriev & Kiselev, 1979).

Voice type High soprano Soprano Mezzo-soprano Tenor Baritone

Frequency of the low singing formant (Hz)

Frequency of the high singing formant (Hz)

Length of the vocal tract (cm)

760–800 700–760 540–600 600–640 540–600

3100–3500 2800–3100 2700–3000 2700–2900 2500–2700

15.3–16.3 16.8–18.5 16.7–18.3 19.0–22.0 21.5–24.0


Vocal tract of singers

The present study attempted to investigate the relationship between the singing voice classification and vocal tract length, vocal tract volume, and formant frequencies using a biologically safe device: an acoustic pharyngometer. An acoustic pharyngometer is a non-invasive device that allows 3-dimensional measurement of the vocal tract. The acoustic pharyngometer makes use of the Acoustic Reflection Technology (ART), which was originally developed for diagnosis of the upper respiratory airway diseases (Xue & Hao, 2006). ART has been widely used to clinically and physiologically examine the upper airway, specifically the nasal passage (Xue & Hao, 2006), the pharyngeal area (Corey, Gungor, Nelson, Fredberg, & Lai, 1997), the endotracheal tube positioning (Huang, Shen, Takahashi, Fukunga, Toga, Takahashi, et al.,1998), and the pharyngeal anatomy of patients with obstructive sleep apnea (Bradley, Brown, Grossman, Zamel, Martinex, Phillipson, et al., 1986; Eckmann, Glassenberg, & Gavriety, 1996), as well as changes in senile vocal tract (Xue & Hao, 2003; Xue, Jiang, Lin, Glassenberg, & Mueller, 1999). ART has been shown to be a valid tool to estimate the human upper airway dimensions (Brooks, Byard, & Fouke, 1989; Brown, Zamel, & Hoffstein, 1986; Eckmann et al., 1996; Kamal, 2004). Its function in a way is similar to a sonar in which an acoustic wave is generated and transmitted along the tube into the airway. A portion of the acoustic wave is reflected back at each point of the discontinuity in the upper airway and is recorded by a microphone attached to the mouthpiece (Kamal, 2004). The other end of the transmitting tube is connected with the computer processor that transforms the wave signal into dimensional values shown on the monitor. The changes in cross-sectional area of the oral cavity can be calculated by comparing the amplitude and temporal changes of the reflected acoustic pulse and the incident pulse. Vocal tract length and volume information can be directly obtained, which makes the ART a more convenient means for vocal tract length and volumetric measure than x-ray technology. ART measurement is rapid and non-invasive, and its accuracy in vocal tract measurements has been well validated with CT scans (D’Urzo, Rubinstein, Lawson, Vassal, Rebuck, Slutsky, et al., 1988; Min & Jiang, 1995) and MRI (Corey et al., 1997; Dang, Honda, & Suzuki, 1994; Jung & Cho, 2004). In the present study, ART was used to investigate the relationship between the dimensional characteristics (length and volume) of vocal tract and singing voice classification.

Method Participants A total of 107 (50 male and 57 female) professional singers were invited to participate in the study. They

3

Table II. Demographic information including means, standard deviation values, and ranges of the participants’ age, height, and weight. Tenors (n ⫽ 27) Age (years) Mean SD Range Height (cm) Mean SD Range Weight (kg) Mean SD Range

24.63 8.46 19–55

Baritones Sopranos Mezzo-sopranos (n ⫽ 23) (n ⫽ 32) (n ⫽ 25) 23.65 8.99 19–62

171.19 173.00 7.98 5.99 153–188 163–188 67.77 15.48 47–125

65.04 6.72 54–76

23.56 4.52 19–34

25.24 9.63 19–53

160.18 6.89 150–176

162.37 6.53 150–176

52.27 7.59 40–69

51.32 6.07 42–67

consisted of 32 sopranos and 25 mezzo-sopranos, and 27 tenors and 23 baritones. Of all participants, 70 were members of the Simon K.Y. Lee Hall choir of the University of Hong Kong (HKU), 23 were members of the Opera Hong Kong (OHK) Chorus, and 10 were active professional soloists recruited from the Opera Society of Hong Kong (OSHK). The type and professional quality of their singing voices were confirmed by the chorus master of OHK Chorus, who possessed more than 30 years of experience in opera performance and instructions. Four other professional soloists were recruited from the Hong Kong Youth Choir (HKYC). Their singing voices were confirmed by the music director of HKYC, who received a Master’s degree of Music in Choral Conducting. The demographic information of the participants including their age, height, and weight data is summarized in Table II. All professional singers had no reported history of any craniofacial abnormalities and no upper airway diseases at the time of experiment. To ensure normal hearing, all of them passed a pure-tone audiometry test bilaterally at the 25 dB level at 250 Hz, 500 Hz, 1000 Hz, 2000 Hz, and 4000 Hz. All professional singers had no vocal abnormalities, such as nodules, polyps, and hoarseness, during the test. Instrumentation The length and volume of the vocal tract of participants were measured by using an acoustic pharyngometer (Eccovision, Pembroke, MA), which is a device that makes use of acoustic reflection technology (ART) to estimate cross-sectional area and length information of an enclosed tube (vocal tract). The acoustic pharyngometer consists of two sound receivers and one sound generator that were mounted on a 30-cm long, 1.89-cm inner diameter tube known as the wave tube, with a microcomputer equipped with A/D conversion ability for processing. To represent the pharyngometric measures a cross-sectional

4

N.Yan et al.

area of the vocal tract as a function of distance from the lips was plotted according to the amplitude and arrival times of the reflected acoustic waves.


Procedure Prior to each recording, the pharyngometer was self-calibrated according to the operator manual. During data collection, each participant was seated upright on a straight-backed chair facing forward. Each participant was instructed to mentally produce the vowel /a/ and breathe out air from his/her mouth to the wave tube of the pharyngometer through the mouthpiece. The corresponding area-distance curve that showed the cross-sectional area of the vocal tract as a function of the distance from the lips to the glottis was subsequently obtained and shown on a computer monitor. This procedure was repeated three times. In addition, to obtain the reference, a similar procedure was repeated once with nose breathing. Upon completion of data collection, a total of four area–distance curves that showed the vocal tract dimension of each participant (thrice with mouth breathing and once with nose breathing) were obtained. Speech and singing voice samples of the participants were recorded by using a high quality stereo voice recorder (WS-200S, Olympus, Allentown/PA, USA). For speech samples, the participants were instructed to produce the vowels /a/, /i/, /u/, /æ/, and /o/; and for singing samples, they sang the “Happy birthday to you” song. The participants were asked to sing with a well-supported voice at modal register, without any adjustment (e.g., vibrato, portamento, staccato, etc.), and changes in vocal register. All recordings were digitized using a sampling rate of 44 kHz and quantization of 16 bits/sample. Data analyses Out of the three cross-sectional area–distance curves obtained using mouth breathing, only one was selected for vocal tract dimension analyses. The selection criteria included: (1) the oral pharyngeal juncture (OPJ) of the mouth-breathing curve best matched with the OPJ of the nose-breathing curve. OPJ is the area where the oral and pharyngeal cavities communicate. Anatomically, it is bounded superiorly by the soft palate, laterally by the anterior faucial pillars, and inferiorly by the tongue dorsum (Zemlin, 1998); and (2) The curve was the most stable: with the least amount of fluctuations in the magnitude caused by changes in airflow (Huang et al., 1998; Xue & Hao, 2006). The selected area– distance curve was then divided into the oral and pharyngeal regions according to the standardized criteria outlined by the manufacturer: (1) the oral region is defined as that from the incisors to the anterior edge of the OPJ, and (2) the pharyngeal region from the OPJ to the opening of the glottis (see

Figure 1). Accordingly, six vocal tract dimensions (oral length, oral volume, pharyngeal length, pharyngeal volume, vocal tract length, and vocal tract volume) were obtained from the selected curves (Xue, Lam, Whitehill, & Samman, 2011). To quantify the vocal tract configuration, the mean and standard deviation values of the six vocal tract dimensions were calculated. Acoustically, speech and singing voice samples were analysed by using Praat (Boersma & Weenink, 2009). Among the five vowel sounds, the vowel /a/ sound was selected for analysis. Meanwhile, two vowels were also selected from the singing samples for analysis, during which the following selection criteria were adopted: (1) the vowel /i/ was extracted from the second vowel of happy; (2) the vowel /æ/ was extracted from the first vowel of the same word. The first three formants (F1, F2, & F3) of the selected vowel sounds were obtained with the autotracking function of the Praat. Reliability measures To assess for inter-examiner reliability, 20 participants were randomly selected for measuring intrarater reliability. The vocal tract dimension of selected participants was re-analysed by the second investigator with using ART. Absolute difference of six vocal tract dimensions between the first time measurements and the second time measurements was obtained (Kamal, 2004). The values of the absolute differences and Pearson correlation coefficients are shown in Table III. It is shown that the two measurements of all six vocal tract dimensions were not significantly different at p ⬍ .01, which indicate that the measurements made by the investigator were both reliable and consistent. Statistical analyses A one-way MANOVA analysis was performed in order to compare each vocal tract parameter against singer voice type. Shapiro-Wilk test, Q-Q plots, and Levene’s test were used to determine normality of data and homogeneity of variances, respectively. Independent-samples t-tests were used to assess for possible gender difference; i.e., tenors vs baritones, and sopranos vs mezzo-sopranos. To supplement the volumetric measurements of the vocal tract, the first three formants were also measured from the selected vowels. For all inferential statistics, a significance level of .05 was used. Results Vocal tract length and volume The mean and the standard deviation values of the six vocal tract parameters (oral volume, oral length, pharyngeal volume, pharyngeal length, vocal tract



5

Figure 1. Area–distance curve of vocal tract dimensions from a pharyngometer. Note: Pharyngeal cavity is calculated by combining oral pharynx and hypo-pharynx (cited from Xue et al., 2011).

volume, and vocal tract length) are summarized in Table IV. For female singers, there was a significant multivariate main effect for singer type (F(4, 51) ⫽ 3.257, p ⫽ .013, λ ⫽ .758, partial ε2 ⫽ .24). Given the significance of the overall test, the univariate main effects were examined. Significant univariate main effects for singer type were obtained for oral length (F(1, 55) ⫽ 7.253, p ⫽ .009, partial ε2 ⫽ .12), total vocal tract length (F(1, 55) ⫽ 8.491, p ⫽ .002, partial ε2 ⫽ .16), oral volume (F(1, 55) ⫽ 5.617, p ⫽ .021, partial ε2 ⫽ .09), and total vocal tract volume (F(1, 55) ⫽ 5.201, p ⫽ .026, partial ε2 ⫽ .09). The sopranos were found to have significantly shorter oral length and total vocal tract length, smaller oral volume, and total vocal tract volume than mezzosopranos significantly. However, no significant difference was found in pharyngeal length and pharyngeal volume of female singers. For male singers, no significant differences were found between higher singing style (tenors) and lower singing style (baritones) in vocal tract dimensions.

First three formant frequencies The mean and standard deviation values of the first three formants of the vowels (/a/, /æ/, and /i/) were summarized in Table V. For female singers, regardless of vowels, no significant differences were found for F1. For F2 values, significant differences were found for /æ/ (t ⫽ 7.057, p ⬍ 0.001) and /i/ (t ⫽ 2.792, p ⫽ .007), whereas no significant differences were found in the vowel /a/ (t ⫽ 1.394, p ⫽ .171). For F3 values, significant differences were present for all vowels (/a/: t ⫽ 5.218, p ⬍ .001; /æ/: t ⫽ 9.348, p⬍.001; and /i/: t ⫽ 3.472, p ⫽ .001). As shown in Table V, higher singing voice types such as sopranos generally exhibited higher formant frequencies than lower singing voice types such as mezzo-sopranos. For male singers, no significant differences were found between higher singing voice types (tenors) and lower singing voice types (baritones) for F1, F2, and F3 of all three vowels (p ⬎ .05).

Table III. Intra-examiner reliability of the six vocal tract measurements. Vocal tract dimensions OL (cm) PL (cm) TL (cm) OV (ml) PV (ml) TV (ml)

First time measurement (mean) 9.89 8.07 17.96 52.99 22.77 75.76

Second time measurement (mean) 9.85 8.08 17.93 52.95 22.72 75.67

Absolute difference

Pearson correlation coefficients (p ⬍ .01)

.04 .01 .03 .04 .04 .08

.991** .985** .983** .999** .999** .989**

OL, Oral length; PL, Pharyngeal length; TL, Total vocal tract length; OV, Oral volume; PV, Pharyngeal volume; TV, Total vocal tract volume. **Significance at p⬍ 0.01.

6

N.Yan et al. Table IV. Mean and standard deviation values of six vocal tract parameters obtained from participants of four singing voices. Singing voice types


OL (cm) PL (cm) TL (cm) OV (ml) PV (ml) TV (ml)

Sopranos (n ⫽ 32)

Mezzo-sopranos (n ⫽ 25)

Tenors (n ⫽ 27)

Baritones (n ⫽ 23)

M

SD

M

SD

p-value

M

SD

M

SD

p-value

9.3 7.5 16.9 47.1 19.4 66.5

.97 1.18 .83 9.40 5.15 9.95

10.0 7.6 17.7 53.1 19.2 72.8

1.03 1.29 1.00 9.72 4.38 10.81

.009* .866 .002* .021* .910 .026*

9.8 8.6 18.5 51.7 25.7 77.5

1.07 1.55 1.45 12.23 8.13 15.67

9.9 8.1 18.0 56.0 25.4 81.4

1.51 1.53 1.09 17.78 7.37 16.20

.936 .235 .185 .325 .882 .391

OL, Oral length; PL, Pharyngeal length; TL, Total vocal tract length; OV, Oral volume; PV, Pharyngeal volume; TV, Total vocal tract volume. *Significance at p ⬍ .05.

Discussion The correlations between vocal tract length and formant frequencies have been discussed in the literature (e.g., Dmitriev & Kiselev, 1979). Generally speaking, the shorter is the vocal tract, the higher are the formant frequencies, which are correlated to higher singing voice types. In the present study, sopranos were found to exhibit significantly shorter total vocal tract length than mezzo-sopranos. Their formant frequencies were also found to be higher than mezzo-sopranos’. This is in line with the findings reported in the literature. Despite the higher formants and shorter vocal tract length, sopranos were found to show a smaller vocal tract volume than mezzo-sopranos. No significant differences in total vocal tract length and the first three formants were found between tenors and baritones. The finding of a shorter vocal tract associated with higher formants is consistent with that reported by Dmitriev and Kiselev (1979). However, our findings of the relationship between vocal tract length and singing voice type did not match with that reported by Dmitriev and Kiselev (1979). For female singers, significant differences were found in both vocal tract length and volume. For male

singers, no significant difference was found in the vocal tract length and volume. Dmitriev and Kiselev (1979) found different formant frequencies associated with tenors and baritones, and such format difference was attributed to the discrepant vocal tract lengths between tenors and baritones. Tenors were found to have shorter overall vocal tract than baritones, with some overlap between the two singing voices (see Table I). This implies that, even with similar vocal tract dimension, two singers could be classified into different singing voices. This appears to suggest that, in addition to vocal tract length, other factors may also affect formant frequencies and the characteristic voice timbre of a professional singer. Dmitriev and Kiselev (1979) suggested professional opera singers used a fixed vocal tract length for all vowels and across the whole voice range of their specific singing voice. Even a singer who changed from low to high notes, his/her vocal tract length remained nearly unchanged. Erickson et al. (2001) hypothesized that singers used timbre transformation when they sang from low notes to high notes. A number of studies suggested that classification of specific singing voices was associated with singers’ vocal fold length (Larsson & Hertegård,

Table V. Mean and standard deviation values of the first three formant frequencies associated with the vowels (/a/, /æ/, /i/) produced by different singers. Singing voice types

Formant frequencies (Hz) /a/

/æ/

/i/

F1 F2 F3 F1 F2 F3 F1 F2 F3

Sopranos (n ⫽ 32)

Mezzo-sopranos (n ⫽ 25)

Tenors (n ⫽ 27)

Baritones (n ⫽ 23)

M

SD

M

SD

M

SD

M

SD

885.54 1481.38 3126.16 749.52 1883.31 3025.36 452.02 2113.03 2885.35

107.47 122.41 161.53 113.96 123.36 141.51 89.18 223.99 174.74

906.43 1420.56 2941.93 789.96 1717.72 2557.05 469.47 1991.37 2732.24

122.84 189.51 155.71 112.63 79.39 202.72 102.08 211.84 122.92

698.82 1206.35 2650.85 585.80 1655.82 2496.97 366.14 1962.86 2578.17

90.69 119.89 220.25 109.54 153.69 150.15 68.27 179.36 171.29

698.93 1208.37 2669.62 598.65 1658.17 2458.69 396.74 1996.68 2585.81

74.52 101.76 189.42 79.27 121.03 110.69 68.66 121.42 139.65



2008; Roers, Murbe, & Sundberg, 2009). Roers et al. (2009) suggested that the morphological differences in singers’ larynx determine vocal fold length and, thus, classification of singing voices. Furthermore, Larsson and Hertegård (2008) found that lower singing voice types had shorter vocal folds length and smaller vocal fold width than those of higher singing voice types. Again, this suggests that, in addition to vocal tract length, other factors may exist that could affect the voice timbre of professional singers and, thus, the singing voice classification. According to vocal tract volumetric data obtained in the present study, sopranos showed a significantly smaller oral volume and total vocal tract volume than mezzo-sopranos, with no significant differences in oral volume, pharyngeal volume, and total vocal tract volume among different male singers. Furthermore, mezzo-sopranos were found to have significantly lower formant frequencies. These findings seem to suggest that, in addition to vocal tract length, vocal tract volume may also affect formants, which in turn create a difference in voice timbre among different singing voices. Recall that professional singers of each singing voice only uses a strictly fixed vocal tract length across the entire voice range (Dmitriev & Kiselev, 1979), and timbre transformation as a singer proceeds from low notes to high notes (Erickson et al., 2001). Should vocal tract length be the sole factor for timbre difference, a timbre transformation would be impossible if professional singers are restricted to use a fixed vocal tract length within their whole voice range. Therefore, other factors that interact with voice timbre should be present. The present finding suggests that vocal tract volume could be a possible factor affecting voice timbre. This is only a speculation. Further investigation examining the relationship between vocal tract volumes and singing voice classification will be needed to confirm this. With regard to formants, oral length and oral volume measures appeared to be systematically correlated with the singer classification. This finding somewhat contradicts with Roers et al. (2009), who reported that the pharyngeal cavity, but not the oral cavity, contributed more to singer classification. The discrepant findings can be attributed to the use of different measurement methods used, as Roers et al. (2009) utilized the x-ray images to estimate the vocal tract length, which included three sections: oral, pharyngeal, and velar regions. The pharyngeal section was defined as the distance from the anterior most point of the laryngeal prominence to the lowermost anterior edge of the second cervical vertebra. The oral length was defined as the distance between the anterior most part of the contour of the eighth molar in the maxilla and the lowermost part of the upper incisor contour (Roers et al., 2009). In the present study, the oral region was defined as the distance from the incisors to the anterior edge of the OPJ and the pharyngeal region as the distance from the OPJ to the opening of the glottis in area-distance curves

7

(see Figure 1). The different definition of oral and pharyngeal regions may explain the discrepant results.

Conclusion The present study found that sopranos showed shorter oral length and total vocal tract length, smaller oral volume, and total vocal tract volume than mezzo-sopranos. Sopranos had generally higher first three vowel formant frequencies than mezzosopranos. These findings suggest that specific vocal tract dimensions and formant frequencies are associated with different professional singing voices. Such results support the argument that the success of professional singing partly depends on the training received, and partly on the vocal tract morphological characteristics of the singers themselves. A limitation of the current study is that the singing voice types of participants were classified by different vocal professionals. The classification criteria of the vocal professionals may be subjective and may vary among vocal pedagogues, which may result in inconsistent classifications of singing voice. This in turn may decrease the external and internal validity of the present study. To maintain consistency of singing voice classification, perhaps future studies should invite only one vocal professional to classify the singing styles of participants. In addition, future studies should recruit equal numbers of singers in different singing voice types. Basses and altos could also be recruited in order to obtain a more complete picture of the relationship between singing voice types and vocal tract dimensions, length, and volume. To further understand the relationship between vocal tract characteristics and singing voice classification, studying the relationship between the timbre transformation and its corresponding changes in vocal tract may be useful. As Erickson et al. (2001) suggested, singers of a certain singing voice type share a similar pattern of timbre transformation. Therefore, it is expected that each singing voice type may demonstrate a characteristic timbre transformation, and thus a specific pattern of vocal tract configuration. Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

References Agren, K., & Sundberg, J. (1976). An acoustic comparison of alto and tenor voices. Journal of Research in Singing, 1, 26–32. American National Standards Institute. (1973). American National Standard psychoacoustical terminology (ANSI S3.20-1973). New York: American National Standards Institute. Berndtsson, G., & Sundberg, J. (1995). Perceptual significance of center frequency of the singer’s formant. Scandinavian Journal of Logopedics and Phoniatrics, 20, 35–41.


8

N.Yan et al.

Boersma, P., & Weenink, D. (2009). Praat: doing phonetics by computer. Available online at: http://www.praat.org/, accessed January 2010. Bradley, T. D., Brown, I. G., Grossman, R. F., Zamel, N., Martinex, D., Phillipson, E. A., et al. (1986). Pharyngeal size in snorers, no snorers, and patients with obstructive sleep apnea. New England Journal of Medicine, 315, 1327–1331. Brooks, L. J., Byard, P. J., Fouke, J. M., & Strohl, K.P. (1986). Reproducibility of measurements of upper airway area by acoustic reflection. Journal of Applied Physiology, 66, 2901–5. Brown, I. G., Zamel, N., & Hoffstein, V. (1986). Pharyngeal crosssectional area in normal men and women. Journal of Applied Physiology, 61, 890–5. Cleveland, T. F. (1977). Acoustic properties of voice timbre types and their influence on voice classification. Journal of the Acoustical Society of America, 61, 1622–1629. Corey, J. P., Gungor, A., Nelson, R., Fredberg , J., & Lai, V. (1997). A comparison of the nasal cross-sectional areas and volumes obtained with acoustic rhinometry and magnetic resonance imaging. Otolaryngology - Head and Neck Surgery, 117, 349–354. Dang, J., Honda, K., & Suzuki, H. (1994). Morphological and acoustical analysis of the nasal and the paranasal cavities. Journal of the Acoustical Society of America, 96, 2088–2100. Detweiler, R. (1994). An investigation of the laryngeal system as the resonance source of the singer’s formant. Journal of Voice, 8, 303–313. Dmitriev, L., & Kiselev, A. (1979). Relationship between the formant structure of different types of singing voices and the dimension of supraglottal cavities. Folia Phoniatrica et Logopaedica, 31, 238–241. D’Urzo, A. D., Rubinstein, I., Lawson, V. G., Vassal, K. P., Rebuck, A. S., Slutsky, A. S., et al. (1988). Comparison of glottic areas measured by acoustic reflection v.s. computerized tomography. Journal of Applied Physiology, 127, 221–226. Eckmann, D. M., Glassenberg, R., & Gavriety, N. (1996). Acoustic reflectometry and endotracheal intubation. Anesthesia and Analgesia, 83, 1084–1089. Erickson, M. L. (2003). Dissimilarity and the classification of female singing voices, a preliminary study. Journal of Voice, 17, 195–206. Erickson, M. L. (2004). The interaction of formant frequency and pitch in the perception of voice category and jaw opening in female singers. Journal of Voice, 18, 24–37. Hollien, H. (1974). On vocal registers. Journal of Phonetics, 2, 125–143. Huang, J., Shen, H., Takahashi, M., Fukunga, T., Toga, H., Takahashi, K., et al. (1998). Pharyngeal cross-sectional area and pharyngeal compliance in normal male speakers and female speakers. Respiration, 65, 458–468. Jung, D. G., & Cho, H. Y. (2004). Predictive value of Kushida Index and acoustic pharyngometry for the evaluation of upper

airway in subjects with or without obstructive sleep apnea. Journal of Korean Medical Sciences, 19, 662–667. Kamal, I. (2004). Test-retest validity of acoustic pharyngometry measurements. Otolaryngology - Head and Neck Surgery, 130, 223–228. Larsson, H., & Hertegård, S. (2008). Vocal fold dimensions in professional opera singers as measured by means of laser triangulation. Journal of Voice, 22, 734–739. Miller, D. G. (2000). Registers in singing: Empirical and systematic studies in the theory of the singing voice. Doctoral Dissertation, University of Groningen, Groningen. Miller, R. (1996). The structure of singing: system and art in vocal technique. New York: Schirmer Books. Min, Y. G., & Jiang, Y. J. (1995). Measurements of cross-sectional area of the nasal cavity by acoustic rhinometry and CT scanning. Laryngoscope, 105, 757–759. Roers, F., Murbe, D., & Sundberg, J. (2009). Voice classification and vocal tract of singers: A study of x-ray images and morphology. Journal of the Acoustical Society of America, 125, 503–512. Seikel, J. A., King, D. W., & Drumright, D. G. (2010). Anatomy and physiology for speech, language, and hearing (4th ed.). NewYork: Delmar Cengage Learning. Sundberg, J. (1973). The source spectrum of professional singing. FoliaPhoniatrica et Logopaedica, 25, 71–90. Sundberg, J. (1974). Articulatory interpretation of the “singing formant”. Journal of the Acoustical Society of America, 55, 838–844. Sundberg, J. (1987). The science of the singing voice. Dekalb, IL: Northern Illinois University Press. Sundberg, J. (1996). Perception of singing voice. In D. Deutsch (Ed.), The psychology of music (2nd ed.). New York, SD: Academic Press. Titze, I. E. (1994). Principles of voice production. Englewood Cliffs, NJ: Prentice-Hall. Vennard, W. (1967). Singing: The mechanism and the technique. New York: Carl Fisher Inc. Xue, A., & Hao, J. G. (2006). Normative standards for vocal tract dimensions by race as measured by acoustic pharyngometry. Journal of Voice, 20, 391–400. Xue, S., & Hao, J. P. (2003). Gender differences in senile changes of human vocal tract and the acoustic correlates of speech production. Journal of Speech, Language, and Hearing Research, 46, 689–701. Xue, S., Jiang, J., Lin, E., Glassenberg, R., & Mueller, P. B. (1999). Age-related changes in human vocal tract configurations and the effects on the speakers’ formant frequencies: A pilot study. Logopedics Phoniatrics Vocology, 13, 245–252. Xue, S. A., Lam, C. W. Y., Whitehill, T. L., & Samman, N. (2011). Effects of class III malocclusion on young male adults’ vocal tract development: A pilot study. Journal of Oral and Maxillofacial Surgery, 69, 845–852. Zemlin, W. R. (1998). Speech and hearing science: Anatomy and physiology. Boston, MA: Allyn and Bacon.