Vocal quality in female classical singers: The role of acoustics, perception and pedagogy Dianna T Kenny Australian Centre for Applied Research in Music Performance (ACARMP), Sydney Conservatorium of Music, University of Sydney, Australia d.kenny@usyd.edu.au - http://www2.fhs.usyd.edu.au/bach/staff/kenny/music.htm Helen F Mitchell Australian Centre for Applied Research in Music Performance (ACARMP), Sydney Conservatorium of Music, University of Sydney, Australia h.mitchell@usyd.edu.au In: K. Maimets-Volt, R. Parncutt, M. Marin & J. Ross (Eds.) Proceedings of the third Conference on Interdisciplinary Musicology (CIM07) Tallinn, Estonia, 15-19 August 2007, http://www-gewi.uni-graz.at/cim07/ Background in music performance and pedagogy Expert listeners are generally reliable in their judgments of overall vocal quality, which have been most closely correlated with tone quality and intonation. Strong correlations have also been observed among different descriptors that assess quality, such as color/warmth, resonance/ ring, clarity/focus, and appropriate vibrato, which indicate that these factors converge on the same underlying construct of overall vocal quality. Pedagogues have recommended many techniques that purportedly produce the desired sound quality. Open throat is one such technique taught in many modern singing studios in the hope that it will produce an desired quality in the sound variously described as even and consistent, balanced and coordinated, round, and warm. Background in psychoacoustics and auditory (music) perception. To date, there have been few perceptual studies that have simultaneously examined acoustic parameters of voice and perceptual preferences of listeners to determine whether there are identifiable acoustic characteristics that are consistently associated with preferred voices. In singing literature, particular LTAS models or exemplars are associated with particular vocal qualities; high-range energy within the LTAS identifies voices that produce a ringing quality or carrying power or amplification over an orchestra which is considered essential to operatic voices. These LTAS patterns have not yet been consistently associated with good or beautiful singing. Aims. In this study, using LTAS, we assessed the relationship between acoustic measurement and perceptual judgment in the identification of perceptually preferred voices and comparing the sound quality in voices using and not using the open throat technique. Method. Fifteen expert judges rated 24 randomly distributed samples with six repeats of six advanced singing students under two conditions: optimal (O), the application of maximal open throat technique; and suboptimal (SO), the application of reduced open throat technique. LTAS were performed on each singing sample, and two conventional assessments of peak energy height [singing power ratio (SPR)] and peak area [energy ratio (ER)], between 0-2 and 2-4 khz were calculated on each LTAS. Perceptual scores, SPR, and ER were rank ordered. Perceptual rankings were then compared with rankings of acoustic measures (SPR and ER) to assess whether these acoustic measurements matched the perceptual rankings of preferred voices. Results. As expected, the audio samples related to the O production were judged by expert listeners to be sung with a better technique than the audio samples related to the SO production. These perceptual rankings, however, did not match the rankings of SPR and ER. Conclusions. SPR and ER were not useful in assessing the vocal quality differences between open throat and reduced open throat singing techniques. These acoustic parameters may distinguish across a wider range of vocal quality but were not sensitive enough to detect changes in the same voices across the different conditions assessed in this study. Implications. Providing a link between acoustic cues and a visual representation of superior quality in LTAS presents a complex challenge. Measures of SPR and ER were never intended to provide vocal quality descriptors, but when the lowest ranked singer on the perceptual ratings achieved the highest rankings in SPR and ER as in this study, these findings indicate that such measures on LTAS may not be able to define vocal quality as perceived by the ear. Future research with LTAS to assess vocal quality should consider alternative measures that are more sensitive to subtle differences in vocal parameters. Because we found that the maximum open throat produced a vocal quality in the singing voice that was judged perceptually by expert listeners to produce the highest overall vocal quality, we therefore tentatively argue that optimal open throat is important to the development of vocal quality.
Are great voices a natural gift or can they be developed by sound pedagogy? What features of the singing voice define a voice of quality? Can these features be measured accurately and more importantly, specifically taught? While a scientific understanding of the voice is not a prerequisite for developing a good voice, it can capture and document vocal pedagogy for future generations. Miller pointed out that for scientific research to be valid and have practical value in the studio, teachers of singing must be involved, knowledgeable and interested. [Their] input, in areas of expertise best understood by voice teachers, is essential (Miller, 1998). There is now both a vast literature on vocal acoustic properties and many unanswered questions about what conclusions can be drawn from such analyses. There are many ways that the voice can be represented acoustically. In current singing research, long term average spectra (LTAS) are widely used to represent timbre and vocal features, both in speech and singing. An LTAS is a measure of the db (decibel) level of the time-average of the power of the acoustic signal at each frequency and it gives an overall impression of an entire excerpt by identifying certain consistent features contained in the sound over time (Jansson & Sundberg, 1975). Understanding patterns in acoustic cues that attract the highest perceptual rankings is needed to identify the most desirable singing voices (Ekholm, Papagiannis, & Chagnon, 1998; Robison, Bounous, & Bailey, 1994; Vurma & Ross, 2000). To date, few scientific studies link acoustic measures with perceptual judgments. Perceptual research indicates that listeners show some degree of reliability and consistency in their overall judgments of good and poor vocal and instrumental performance (Ekholm et al., 1998; Geringer & Madsen, 1998; Saunders & Holahan, 1997; Smith, 2004; Wapnick, Flowers, Alegant, & Jasinskas, 1993) and more specifically in their assessment of excellence in overall voice quality (Kenny & Mitchell, 2006; Stanley, Brooker, & Gilbert, 2002; Wapnick & Ekholm, 1997). Establishing criteria specific rating schemes may guide the listening task. However, Ekholm, Papagiannis et al. (1998) found high correlations between perceptual criteria and overall judgment of quality. Using a number of criteria, rather than a single overall judgment may be redundant and not provide additional information as that provided from the unitary measure of overall quality (Thompson, Diamond, & Balkwill, 1998). Because responses to a musical performance are subjective, reliance on panels of judges with extensive knowledge and experience is necessary. (Davidson & Da Costa Coimbra, 2001; Johnson, 1997) Musical pedagogy has evolved through an oral, non-scientific tradition. Vocal pedagogues must assess each individual voice and devise a technical and aesthetic program to improve its basic sound. One technical component of many modern singing studios is open throat. Mitchell et al. (2003) reported that singing pedagogues actively identify open throat as a critically important technique in classical singing and assert that it produces an audible quality in the sound, variously described as even and consistent, balanced and coordinated, round and warm. In subsequent studies, they compared the use of open throat technique (O) to a reduction of the technique described as sub-optimal (SO) in the same advanced singing students and found that experts could identify the technique in 83% of samples (Mitchell & Kenny, 2004, 2006). While there has been major progress in the use of LTAS in singing research, the study reported here is the first to our knowledge to compare perceptual responses with LTAS exemplars of operatic/classical voices. Participants Method Listeners were 15 experienced singing pedagogues, 12 females and 3 males aged between 37 and 76 with a mean of 54 years. Six participants had a postgraduate qualification in singing, five had a diploma of music or singing, and three a bachelor degree in music. One cited extensive international performing experience as their qualification in singing. All had taught singing for 4 to 40 years, with an average for 20 years. Thirteen of fifteen participants taught singers in a Conservatorium of music. 2
Overall, participants singing studios comprised an average of 39.5% of operatic students and 36.7% classical students. For 11 pedagogues, the majority of their studio comprised these two genres. Six pedagogues also taught 20% of musical theatre students in their studio. Eleven taught a proportion of international and national level singers, nine at big city or regional/touring and eight at local community level. Singing stimuli Six advanced singing students sang two song excerpts, Mozart song Ridente la Calma and Schubert lied Du bist die Ruh under two conditions: Optimal, and sub-optimal. Optimal (O) condition was necessary to provide a sound quality with the best technique that they could. This involved the maximal use of open throat. Sub-optimal (SO) condition involved the use of a reduced (open throat) technique but still with an acceptable singing technique and without consciously altering any other aspect of their technique. The acoustic signal was recorded digitally (Behringer Ultragain preamplifier/marantz CDR 630) via a high-quality microphone (AKG C-477) positioned on a head boom a constant 7 cm distance from the singer s lips. Pink noise was played from a speaker (Bose Lifestyle) and the microphone of a sound level meter (Rion NL-06 SPL) was coupled to the AKG microphone for calibration of absolute sound pressure levels (SPL). The audio recordings of the singers were digitally extracted using Audiograbber software (www.audiograbber.com) on a standard PC computer to wav audio format in stereo at 41000 Hz and 16 bit sample rate. The files were then reopened in Cool Edit Pro 1.2. The first 7 bars of Mozart and bars 54 to 60 of the Schubert musical tasks were used for the perceptual study. Each consisted of an entire musical phrase lasting around 20 seconds. Acoustic analysis The audio recordings were digitised at 16 khz sample rate using Phog Version 2.0 (Hitech, Sweden) software and analysed in Soundswell Version 4.0 (Hitech, Sweden) using A- weighting for SPL pink noise calibration and subject SPL measurement. Long term average spectra (LTAS) analyses (bandwidth 300 Hz) were performed on the 24 musical task files (Nordenberg & Sundberg, 2003; White, 1998). A conventional way of assessing LTAS is to reduce the information it contains to a single meaningful number, by computing the ratio of energies in a low and a high frequency band (Omori, Kacker, Carroll, Riley, & Blaugrund, 1996; Thorpe, Cala, Chapman, & Davis, 2001). In singing, measures of spectra compare energy peak height [singing power ratio (SPR)] and peak area [energy ratio (ER)] between 0-2 khz and 2-4 khz. The difference between the height of the major peaks between 0-2 and 2-4 khz (Omori et al., 1996) quantifies the relative energy between 2-4 khz although it does not account for the shape of these energy peaks. Assessing the areas under the LTAS curve at 0-2 and 2-4 khz, rather than the highest peaks of energy, also identifies the areas in which the energy is reinforced or reduced (Thorpe et al., 2001). Both SPR and ER enable effective comparison between individual singers and between groups of singers. Procedure The perceptual test was conducted in a quiet environment and samples were played from a Sony CD Walkman (DEJ885W) via circumaural closed-back stereo monitoring headphones (Sennheiser HD 270). Prior to presentation of stimuli, participants were given information on the two singing conditions, O and SO, and were presented with the musical score of each musical task. They were asked to judge whether or not the singer was using open throat technique in each sample. They were asked to rate each sample on a 10-point scale. Judges who assessed singers were not the singers pedagogues and did not know any of their identities. Listeners assessed 30 tracks [6 singers x 2 conditions, (O and SO) x 2 musical tasks] in random order, including six repeats, also selected at random. 3
Results Intraclass correlations with a two-way random effects model of absolute agreement (ICC (2,1) ) compared the scores of the repeated samples with the scores of the original sample. For the O condition, ICC =.7843 [CI 95%.4822 -.9613] F = 9.52, p <.0001. For SO, ICC =.7579 [CI 95%.4338 -.9558] F = 8.60, p <.0001. Listeners matched their exact rating in the first rendition and the repeat in nearly 5 of 6 cases. Therefore, judges were reliable in their absolute ratings. Overall means of score by condition indicated a significant reduction in SO compared to O in both musical tasks. For mean score, there was a main effect for condition (F (1, 5) = 111.872, p =.000), but not for musical task (F (1, 5) = 1.497, p=.276); that is, ratings were significantly higher for O than for SO in both the Mozart and Schubert tasks. There was no interaction between musical task and condition (F (1, 5) =.355, p =.577). Table 1 presents the rankings of each singer and the respective musical task and experimental condition, ordered by perceptual rank from highest to lowest. Pearson s correlation coefficients were calculated for each of the three dependent measures (perceptual rating, SPR and ER). While there is a very highly significant relationship between ER and SPR, as expected (.960* p=.000) there is no relationship between either of these acoustic measures and perceptual judgments (ER, 0.064, p=.767; SPR, 0.101 p=.638). Intraclass correlations with a two-way random effects model of absolute agreement (ICC (2,1) ) compared the scores of the repeated samples with the scores of the original sample. For the O condition, ICC =.7843 [CI 95%.4822 -.9613] F = 9.52, p <.0001. For SO, ICC =.7579 [CI 95%.4338 -.9558] F = 8.60, p <.0001. Listeners matched their exact rating in the first rendition and the repeat in nearly 5 of 6 cases. Therefore, judges were reliable in their absolute ratings. Spearman s rho correlation coefficients were calculated for judges rankings of samples and singers on each of the dependent measures. These data indicate that there was no relationship between perceptual ratings of samples or singers based on ranked ER (- 0.006, p=.977) and SPR (0.308, p=.143). Percept Rank SPR Rank ER Rank Singer Task Cond 1 16 9 5 S O 2 13 3 5 M O 3 10 10 1 M O 4 5 2 3 M O 5 9 11 1 S O 6 21 20 4 S O 7 6 13 4 M O 8 23 23 2 S O 9 19 19 3 S O 10 8 6 2 M O 11 11 14 6 S O 12 4 4 5 S SO 13 3 12 1 M SO 14 24 24 2 S SO 15 15 8 6 M O 16 7 7 5 M SO 17 20 21 3 S SO 18 18 18 6 S SO 19 12 16 1 S SO 20 1 5 3 M SO 21 22 22 4 S SO 22 17 15 2 M SO 23 14 17 4 M SO 24 2 1 6 M SO Table 1. Singer, task, condition and rankings for perceptual score, singing power ratio (SPR) and energy ratio (ER), sorted by perceptual score ranking from highest to lowest. Figures 1a-d present LTAS performed on selected samples used in the perceptual study. LTAS may provide acoustical interpretation of the perceptual ratings or conversely, acoustic cues that influenced listeners ratings may be evident in the LTAS. Table 2 presents judges qualitative comments. 4
Figure 1. LTAS of highest (a), 3= (b), mid (c) and lowest (d) perceptually ranked singers. Legends correspond to task, Mozart (M) or Schubert (S), and singer number. Rank Positive Negative Highest "Mellow and appealing overall quality" "Operatic, secure, well controlled" "Warm, rich and consistent" "Less resonant, less bright" "Too far back" "Exaggerated darkness of tone" Second highest Mid Lowest "focused, warm overtones" "chiaroscuro tone, generous tone with warmth" "good foundation tone, voice is still building" "still solid core" "more ring, twang, brighter vowels" "too 'covered' - too far back" "lacks freedom" "lacks vibrancy and warmth" "straight, white" "no warmth or overtones, straight sound, no vibrato" Table 2. Exemplars of qualitative positive and negative descriptions of the sound quality by singer rank. Singer 5 was rated highest overall, in both Mozart and Schubert tasks (7.60, 7.07), and singers 1 and 3 were ranked third with a score of 6.93. For the O condition, the mean scores were 6.34 for Mozart and 6.61 for Schubert. The singers O condition scores ranged from 5.40 to 7.60. For SO, singer 5 was rated highest for Schubert with a score of 5.40. For SO, the mean scores were 3.81 for Mozart and 4.31 for Schubert. Singers SO scores ranged from 2.67 to 5.40. Singers O scores consistently scored higher rating than SO. Discussion In this study, listeners were consistent in their ratings of overall quality, producing exactly the same score, on average, for five of the six repeated samples. There were no statistically significant differences between listeners ratings of the two musical tasks. As expected, the audio samples of "optimal" production were judged by expert listeners to be sung with a better technique than the audio samples of sub-optimal production. Overall preferences for particular singers were also similar and judges ranked their favourite singers (5, 1 and 3) consistently 5
higher. Of these, singer 5 scored the highest overall ratings. High scoring singers were praised for their overall vocal quality, or combinations of positive qualities that resulted in a balanced sound (Mitchell et al., 2003). Judges recognised an overall good quality in the midscoring voices, but identified technical flaws that impacted on vocal production. In general, judges were more likely to comment on faulty vocal production or technique, particularly in those ranked lowest. Negative comments across all singers were more descriptive and focussed on technical suggestions for improvement in the sound. Conversely, those ranked highly attracted little comment. It seems that a beautiful voice is difficult to describe. We examined LTAS measurements with respect to the perceptual ratings of singers. Acoustic measurements were ranked for vocal quality, specifically, carrying power after the work of (Barnes, Davis, Oates, & Chapman, 2004; Omori et al., 1996; Thorpe et al., 2001). As expected, there was a strong relationship between the two acoustic measures, but there was no relationship between perceptual rankings of vocal beauty and acoustic rankings of vocal quality. Acoustic measures such as LTAS were never intended to provide vocal quality descriptors (Bartholomew, 1934), but when the lowest ranked singer on the perceptual ratings achieved the highest rankings in SPR and ER as in this study, such findings indicate that a simple analysis such as LTAS cannot define vocal quality as perceived by ear. The LTAS of O and SO conditions produced in this study were both different to LTAS produced in other studies examining speech quality (Barrichelo, Heuer, Dean, & Sataloff, 2001; Löfqvist, 1986) pop singing (Borch & Sundberg, 2002) or country singing (Cleveland, Sundberg, & Stone, 2001). Therefore, the energy distribution of LTAS in this study may represent an intrinsic component of operatic or classical sound. We therefore argue that energy distribution above 2 khz indicates a fundamental quality of classical and operatic singing but is not necessarily an indicator of overall vocal quality. Measures of SPR and ER may concur more closely with perceptual studies of vocal projection, carrying power or potential amplification over an orchestra. Vocal quality appears to be more than a complex combination of acoustical parameters. To date, no single objective evaluation captures or characterizes vocal quality in a systematic way (Omori et al., 1996; Thorpe et al., 2001). We have demonstrated that acoustic analyses such as LTAS do not reliably match perceptual judgments by expert listeners and therefore cannot be used to define or predict vocal quality. We recommend that any future acoustic analyses or visual representations of voice must emulate the human ear. Applied research in voice is in need of better perceptual and acoustic measures to define the singing voice. While there appear to be some acoustic cues in the singing samples in this study that were associated with listener preferences, there is much work to be done in characterizing the voice visually using LTAS that provides a better association with perceptual rankings of expert listeners. Future research could examine acoustically many exemplars of vocal quality in a further attempt to find communalities in the acoustic profiles of such voices. Acknowledgments. We thank Ms Maree Ryan for her pedagogical advice and support. We also thank Mr Peter Thomas and Dr Densil Cabrera for their advice on acoustical matters and the fifteen listeners and six singers for their willing participation in the project. References Barnes, J. J., Davis, P. J., Oates, J., & Chapman, J. (2004). The relationship between professional operatic soprano voice and high range spectral energy. Journal of the Acoustical Society of America, 116(1), 530 538. Barrichelo, V. M. O., Heuer, R. J., Dean, C. M., & Sataloff, R. T. (2001). Comparison of singer's formant, speaker's ring and LTA spectrum among classical singers and untrained normal speakers. Journal of Voice, 15(3), 344-350. Bartholomew, W. (1934). A physical definition of "good voice quality" in the male voice. Journal of the Acoustical Society of America, 6, 25-33. 6
Borch, D. Z., & Sundberg, J. (2002). Spectral distribution of solo voice and accompaniment in pop music. Logopedics Phoniatrics Vocology, 27, 37-41. Cleveland, T. F., Sundberg, J., & Stone, R. E. (2001). Long-Term-Average-Spectrum characteristics of country singers during speaking and singing. Journal of Voice, 15 (1), 54-60. Davidson, J. W., & Da Costa Coimbra, D. (2001). Investigating performance evaluation by assessors of singers in a music college setting. Musicae Scientiae, 5(1), 33-53. Ekholm, E., Papagiannis, G. C., & Chagnon, F. P. (1998). Relating objective measurements to expert evaluation of voice quality in Western classical singing: Critical perceptual parameters. Journal of Voice, 12 (2), 182-196. Geringer, J. M., & Madsen, C. K. (1998). Musicians' ratings of good versus bad vocal and string performances. Journal of Research in Music Education, 46(4), 522-534. Jansson, E. V., & Sundberg, J. (1975). Long-Time- Average-Spectra applied to analysis of music. Part I: Method and general applications. Acustica, 34, 15-19. Johnson, P. (1997). Performance as experience: The problem of assessment criteria. British Journal of Music Education, 14(3), 271-282. Kenny, D. T., & Mitchell, H. F. (2006). Acoustic and perceptual appraisal of vocal gestures in the female classical voice. Journal of Voice, 20 (1), 55-70. Löfqvist, A. (1986). The long-time-average spectrum as a tool in voice research. Journal of Phonetics, 14, 471-475. Miller, R. (1998). The singing teacher in the age of voice science. In R. T. Sataloff (Ed.), Vocal Health and Pedagogy (pp. 297-300). San Diego, CA: Singular. Mitchell, H. F., & Kenny, D. T. (2004). The impact of "open throat" technique on vibrato rate, extent and onset in classical singing. Logopedics Phoniatrics Vocology, 29(4), 171-182. Mitchell, H. F., & Kenny, D. T. (2006). Can experts identify open throat technique as a perceptual phenomenon? Musicae Scientiae, X(1), 33-58. Mitchell, H. F., Kenny, D. T., Ryan, M., & Davis, P. J. (2003). Defining open throat through content analysis of experts' pedagogical practices. Logopedics Phoniatrics Vocology, 28(4), 167-180. Nordenberg, M., & Sundberg, J. (2003). Effect on LTAS of vocal loudness variation. TMH- QPSR, 45, 93-100. Omori, K., Kacker, A., Carroll, L. M., Riley, W. D., & Blaugrund, S. M. (1996). Singing power ratio: Quantitative evaluation of singing voice quality. Journal of Voice, 10(3), 228-235. Robison, C. W., Bounous, B., & Bailey, R. (1994). Vocal Beauty: A study proposing its acoustical definition and relevant causes in classical baritones and female belt singers. Journal of Singing, 51, 19-30. Saunders, T. C., & Holahan, J. M. (1997). Criteriaspecific rating scales in the evaluation of High School instrumental performance. Journal of Research in Music Education, 45(2), 259-272. Smith, B. P. (2004). Five judges' evaluation of audiotaped string performance in international competition. Bulletin of the Council for Research in Music Education(160), 61-69. Stanley, M., Brooker, R., & Gilbert, R. (2002). Examiner perceptions of using criteria in music performance assessment. Research Studies in Music Education, 18, 43-52. Thompson, W. F., Diamond, P. C. T., & Balkwill, L.- L. (1998). The adjudication of six performances of a Chopin etude: A study of expert knowledge. Psychology of Music, 26, 154-174. Thorpe, C., Cala, S., Chapman, J., & Davis, P. (2001). Patterns of breath support in projection of the singing voice. Journal of Voice, 15(1), 86-104. Vurma, A., & Ross, J. (2000). Priorities in voice training: Carrying power or tone quality. Musicae scientiae, 4(1), 75-93. Wapnick, J., & Ekholm, E. (1997). Expert consensus in solo voice performance evaluation. Journal of Voice, 11(4), 429-436. Wapnick, J., Flowers, P., Alegant, M., & Jasinskas, L. (1993). Consistency in piano performance evaluation. Journal of Research in Music Education, 41(4), 282-292. White, P. (1998). A study of the effects of vocal intensity variation on children's voices using long-term average spectrum (LTAS) analysis. Logopedics Phoniatrics Vocology, 23, 111-120. 7