The Singing Power Ratio as an Objective Measure of Singing Voice Quality in Untrained Talented and Nontalented Singers

*Christopher Watts, †Kathryn Barnes-Burroughs, ‡Julie Estis, and ‡Debra Blanton

*Harrisonburg, Virginia, †Lubbock, Texas, and ‡Mobile, Alabama

Summary: A growing body of contemporary research has investigated differ- ences between trained and untrained singing voices. However, few studies have separated untrained singers into those who do and do not express abilities related to singing talent, including accurate pitch control and production of a pleasant timbre (voice quality). This investigation studied measures of the singing power ratio (SPR), which is a quantitative measure of the resonant quality of the singing voice. SPR reflects the amplification or suppression in the vocal tract of the harmonics produced by the sound source. This measure was acquired from the voices of untrained talented and nontalented singers as a means to objectively investigate voice quality differences. Measures of SPR were acquired from vocal samples with fast Fourier transform (FFT) power spectra to analyze the amplitude level of the partials in the acoustic spectrum. Long-term average spectra (LTAS) were also analyzed. Results indicated significant differences in SPR between groups, which suggest that vocal tract resonance, and its effect on perceived vocal timbre or quality, may be an important variable related to the perception of singing talent. LTAS confirmed group differences in the tuning of vocal tract harmonics. Key Words: Singing talent—Singing ability—Resonance—Singing power ratio—Phonatory control.

Accepted for publication December 9, 2004. INTRODUCTION From the *Department of Communication Sciences & Disor- ders, James Madison University, Harrisonburg, Virginia; The facility to express a skill in the performance †School of Music, Texas Tech University, Lubbock, Texas; and of a task is often credited to talent. Talent can ‡Department of Speech Pathology & Audiology, University of be thought of as a special natural ability that has South Alabama, Mobile, Alabama. Address correspondence and reprint requests to Christopher the potential to lead to a capacity for achievement R. Watts, Department of Communication Sciences & Disorders, or success. Anecdotally, physiological talent is com- James Madison University, MSC 4304, Harrisonburg, VA 22807. monly recognized, most notably, in sports perfor- E-mail: [email protected] mance. Singing talent is also recognized, such as Journal of Voice, Vol. 20, No. 1, pp. 82–88 when judges critique the degree of talent expressed 0892-1997/$32.00 ą 2006 The Voice Foundation by a person for making decisions regarding awards or doi:10.1016/j.jvoice.2004.12.003 admittance to specialized training programs. Singing

82 SINGING POWER RATIO 83 skill, which comes from talent and refers to the Both factors can be measured objectively or experi- degree of singing excellence, can be honed even enced perceptually by a listener. Intonation is influ- more and perfected with training and practice. enced by pitch control abilities. Timbre, or voice Singing involves both physiological and artistic quality, is influenced by the supraglottal configura- components. Physiologically, singing talent can be tion of the vocal tract and its resonant effects on the defined as the special natural ability to coordinate and voice source. control the musculature and structures of the voice It is possible to objectively investigate pitch con- organ to produce musical modulations of the trol and voice quality in talented and nontalented voice where the sounds vary over a wide range of persons via acoustic analyses of vocal productions. frequencies and are in tune with each other, or where Study of these variables may provide evidence such sounds are melodious.1 Singing talent can toward understanding the physiological capabilities be inferred from listening to a person sing. In fact, that distinguish one group from the other. The many professional voice teachers have frequent ex- singing power ratio (SPR) has objectively studied perience making these types of judgments. It is likely singing voice quality in both trained and un- that singing talent is expressed along a continuum trained singers.9,10 This measure is an indicator of so that there are different degrees of ability. How- resonant tuning in the vocal tract, and evidence has ever, this fact is not necessarily taken into consider- suggested that SPR represents the acoustic charac- ation universally in many real-world contexts. teristics of singing voice quality.9 SPR is calculated Specifically, binary judgments regarding singing by measuring the ratio of the peak intensities be- talent (eg, present vs. not present; at a required level tween the 2–4-kHz and 0–2-kHz frequency bands in vs. not at a required level) are routinely applied by the context of sustained vowels or vocalic segments experienced judges to persons competing for admit- in sung/spoken samples. Lower SPR measures indi- tance to college music programs, for voice scholar- cate greater energy in the higher harmonics, a con- ships, and in other competitions. dition that has been said to enhance the richness and Little empirical data are available that investigate, perception of “ring” in the voice, and thus it influ- physiologically, why some people, who are un- ences the perception of voice quality in a positive trained, express singing talent, whereas others do way.10 This measure has been found to differenti- not. Goetze et al2 suggested several variables that ate the singing voices of trained and untrained sing- are requisite for accurate singing (accurate singing, ers, as well as spoken and sung productions.9,10 or singing in tune, is one variable thought to be Judgments of singing talent are usually exclu- highly important for the expression of singing sively subjective. However, perceptual experience talent), which included the ability to discriminate of voiced sound is influenced by the spectral rela- between pitches, vocalize over a wide range of tionship of the harmonics in the sound reaching the pitches, the ability to monitor vocal pitch, and a ear. If judges are asked to include timbre as a factor motivation to sing.2 The importance of pitch control in rating singing talent, then analysis of the voice for accurate singing has also been reported by other spectrum can assist in determining the influence of authors.3–7 timbre on those judgments. SPR is one type of mea- To investigate which perceptual factors were most sure that we can use toward this goal. Another important for judgments of singing talent, Watts measure from which voice quality can be inferred et al8 conducted a large survey of professional voice is long-term average spectra (LTAS). LTAS mea- teachers. Over 1000 participants responded to ques- sures allow inferences regarding the sound source tions that queried what factors were most important spectrum and can be calculated from connected in judging whether an untrained voice expressed speech samples, whereas SPR focuses more on the singing talent. According to these experts, the three influence of resonance in the supraglottal vocal tact mostimportantvariableswereintonation, timbre,and and its measurement is restricted to the context of musicality. The first two variables, intonation and vocalic productions. LTAS averages spectral fea- timbre, related to physiological singing abilities (that tures of the speech signal over time, from which is, pitch control and voice quality, respectively). aspects of voice quality (timbre) can be inferred via

Journal of Voice, Vol. 20, No. 1, 2006 CHRISTOPHER WATTS ET AL84 calculation of spectral slope or tilt.11,12 Spectral tilt Only data from 33 participants were included in represents the drop in energy from lower frequency the final analysis. Six subjects either did not meet harmonics to the higher frequency harmonics in a inclusion criteria or were not rated unanimously by spectrum. LTAS is averaged over time, and it judges as falling into one of the experimental sub- analyzes longer utterances that include speech. groups (see below). All participants had normal Therefore, LTAS is a useful tool for analysis of voice hearing at 20-dB HL for 500 through 4000 Hz, quality in sung passages. no history of chronic vocal pathology, no previous This study is one of a series of investigations voice therapy, no history of drug or alcohol abuse, focusing on objective measures of phonatory, per- no history of smoking, and reported no allergies or ceptual, and resonant tuning abilities in untrained voice problems at the time of testing. In addition, talented and nontalented singers, where perceptual no subjects had received one-on-one voice training judgments of singing talent are based on intonation from either a professional or amateur teacher. accuracy and voice timbre (eg, voice quality). The To assign participants into experimental groups, goal of this line of research is to collect physiological recordings of a song stimulus produced by each evidence, both indirectly and directly, for the pur- participant were presented to two trained profes- pose of explaining differences in singing talent in sional teachers of singing. Both judges were women, untrained persons. The specific research questions of possessed graduate degrees in voice performance, this study are as follows: (1) Do untrained persons were voice faculty in a university setting with experi- with talented singing voices resonate the source ence at rating singing talent, and had over 5 years spectrum (eg, modify the energy relationships of of professional pedagogical experience. Each judge harmonics produced by the vibrating vocal folds) listened to the recordings independently. The identi- differently when singing compared with untrained ties of the singers were not disclosed to the judges. nontalented persons? (2) Do untrained persons with The judges were provided the singing samples on singing talent demonstrate a different degree of spec- a compact disk and asked to judge whether the tral tilt when singing compared with untrained non- person singing expressed singing talent. The judges talented persons? It is hypothesized that persons were instructed to make binary judgments (present / judged to express a talented singing voice will mani- not present) of perceived talent, and to base their fest lower SPR measures and less spectral tilt, both judgments solely on the intonation (pitch accuracy) of which would be indicative of greater energy in the higher harmonics, compared with persons judged and timbre (voice quality) of the voice during sing- to lack singing talent. ing. Each judge reported understanding of the in- structions and indicated that they felt comfortable, and had experience, with those two variables as criteria for rating vocal talent. METHOD Only subjects who were rated unanimously by Participants both judges as either expressing or not expressing Participants included 39 persons recruited from singing talent were assigned to one of the two the undergraduate and graduate populations of groups. Any subject who was rated differently by speech and hearing majors at the University of South the two judges was excluded from the study. Out Alabama to participate in an ongoing line of investi- of the 39 initial participants, judges agreed on 34 gation into singing accuracy and talent. Each partic- ratings (87%). Twelve subjects were rated unani- ipant had the potential to be assigned to one of two mously as expressing accurate intonation and a voice groups, which, for the purposes of this study, were quality pleasant enough to be characterized as tal- labeled as the talented and nontalented groups. ented, whereas 22 participants were rated unani- Group assignment was based on judgments made mously as nontalented. One subject rated as by professional voice teachers, as described below. nontalented was found to have a history of neurolog- Participants were all women, and they ranged in age ical problems and excluded from the study. To assess from 20 to 35 years (mean ϭ 23 years). reliability of judgments, a third professional voice

Journal of Voice, Vol. 20, No. 1, 2006 SINGING POWER RATIO 85 teacher listened to 13 randomly selected (33%) stim- of 1024 points and a bandwidth of 8000 Hz. For uli sung by the pool of participants who were rated each vocalic segment, energy peaks were calculated unanimously by the two original judges. There was between 0 and 4000 Hz. SPR was then calculated by 100% agreement (6 talented; 7 nontalented) between subtracting the amplitude of the strongest partial the ratings of the third judge and those of the first two, between 2 and 4 kHz from the level of the strongest which suggests that the original ratings of talent/ partial between 0 and 2 kHz. This measure has been nontalent based on intonation and timbre were reported to reflect the “ring” in the voice, and it reliable. corresponds to the resonant quality of the singing voice.9,10 Procedures LTAS were also calculated by analyzing the stim- Before testing, participants read and signed a con- ulus songs via the CSL. LTAS were calculated in sent form, completed questionnaires, and underwent this study as a measure of spectral tilt of the fre- audiometric screening to ensure hearing was within quencies between 0 and 8000 Hz, in increments of the required limits for the study. Participants were 500 Hz, for comparison with the SPR measures. then asked to sing the first stanza of the song LTAS were calculated from the entire duration of “America the Beautiful” in their best singing voice at each song sung by the subjects, and the harmonic comfortable pitch and loudness levels. To facilitate peaks were averaged across each group for in- the best singing sample from participants, they tergroup comparison. Because loudness variations were given a minimum of 5 minutes for vocal warm- can affectthe intensity in higher spectrumpartials, the up, including time to practice the stimulus song. amplitude values of the spectral peaks were normal- Every participant reported that they were familiar ized to control for loudness variations between sub- with the song. If any participant was unfamiliar with jects. This process was accomplished by assigning the tasks for vocal warm-up, they were given instruc- the intensity of the strongest partial a value of 1 tion. Finally, participants were informed that no one and each subsequent partial a proportional value could hear them while singing, as they were to compared with this peak intensity. be alone in a sound-treated booth during production. An independent samples t test was applied to the Recording took place in a double-walled sound SPR data to investigate differences between the two booth with the subject in isolation, and sung samples experimental groups, with the significance level set were recorded digitally at a sampling rate of 44.1 at P Յ 0.05. Averaged LTAS data were examined kHz onto an Alesis Masterlink (Alesis West Coast, graphically to assess spectral profiles in the two Los Angeles, CA) digital recorder via a direct line groups, and compared with the SPR results. input from an AKG Acoustics (AKG USA, Nashville, TN) head-mounted microphone posi- tioned 3 cm off-center at the left corner of the mouth of the subject. These samples placed each participant RESULTS into their respective group based on the perceptual Preliminary comparisons of subject performance judgments of the professional voice teachers (see revealed apparent dissimilarities in performance above), and for later spectral analysis. between the two groups. Table 1 displays measures of SPR in subjects of both groups. Collectively, the Data calculation and analyses talented group manifested an SPR that was approxi- Two measures were calculated to compare har- mately 8 dB less than the nontalented group. Almost monic energy in the two groups of subjects: SPR half of the nontalented subjects manifested SPR and LTAS. For SPR measures, fast Fourier transform values over 30 dB, whereas only one talented subject (FFT) power spectra were calculated from selected produced a value that high. The variance was very vocalic segments (five total) in the recordings of the similar (6.9 dB for talented, 5.7 for nontalented) stimulus song with the Kay Elemetrics Computer- between the two experimental groups. The data war- ized Speech Lab (CSL, model 4400; Kay Elemetrics ranted more investigation to test for group Corporation, Lincoln Park, NJ), with a time window differences.

Journal of Voice, Vol. 20, No. 1, 2006 CHRISTOPHER WATTS ET AL86

TABLE 1. Individual Means and Group Means With Standard Deviations on Measurements of Formant

Power Difference (SPR) in dBSPL 35.00 Group Subject Talented Nontalented 1 19.3 27.8 30.00 2 20.6 24.0 3 18.9 31.1 4 22.0 29.4 5 17.9 26.3 25.00 6 18.9 29.5

7 34.1 33.6 Singing Power Ratio (dB) 8 24.6 26.6 9 26.5 28.6 20.00 10 18.8 31.8 11 21.2 33.6 12 28.5 36.0 Talented Nontalented 13 – 29.9 group 14 – 33.6 15 – 31.5 FIGURE 1. Box plot of averaged SPR values (in dBSPL) 16 – 32.8 for subjects in the talented and nontalented groups (variability 17 – 30.9 illustrated by vertical lines). 18 – 28.5 19 – 28.6 20 – 34.7 21 – 37.0 95% CI ϭ 9.92–6.04). SPR scores were signifi- Group avg. 22.6 (6.9) 30.7 (5.7) cantly lower for the talented group compared with the nontalented group. Lower SPR scores corre- spond to an increase in perceived vocal “ring” and Figure 1 illustrates a box plot representing aver- the perception of enhanced voice quality (timbre) aged SPR values for subjects in both experimental because of the greater energy in the higher harmon- groups. Again, higher values of SPR represented ics. The level of significance and confidence interval less energy in the higher harmonics (above 2000 suggested a robust finding. Hz) in ratio to the energy in the lower harmonics Figure 2 illustrates averaged LTAS results for each (below 2000 Hz). Although the variability of the group. As can be seen, the talented subjects dis- talented group seems to be high, the superior vertical played lower spectral tilt, which means that the line on the talented plot represents only one subject higher harmonics were more dominant in the spec- in this group who produced an SPR value that was trum compared with the nontalented group, where more than one standard deviation from the mean the fundamental and lower harmonics were more (talented subject 7). Conversely, only one subject dominant. The talented group demonstrated a no- in the nontalented group (nontalented subject 2) pro- ticeable rise in energy throughout the 2500-Hz to duced a SPR value less than one standard deviation 4000-Hz frequency band. These measures agreed from the mean, as illustrated by the inferior vertical with the results from the SPR data. line on the nontalented plot. Collectively, the separa- tion in means between the two groups is apparent. The Levene test for equality of variances was DISCUSSION above 0.05, so the data were analyzed assuming equal group variances. Results from the t test re- This study compared measures of the SPR, which vealed a significant difference between the two reflected harmonic tuning of the source spectrum by groups with regard to SPR (t ϭ 8.13, P Ͻ 0.001.; the superior vocal tract, in untrained talented and

Journal of Voice, Vol. 20, No. 1, 2006 SINGING POWER RATIO 87

1 of investigating physiological abilities during sing- 0.9 ing. Harmonic tuning of the vocal tract is partially 0.8 dependent on the given anatomical structure of a

harmonic) 0.7

st person, but also dependent on the physiological abil- 0.6 Talented ity to manipulate the vocal tract to tune the voice 0.5 Nontalented spectrum for the production of a target voice quality. 0.4 Judgment related to the presence or absence of 0.3 expressed singing talent is also influenced by pitch 0.2 control.8 Previous studies have found that subjects 0.1

dB (re: amplitude of 1 judged to produce talented singing voices show 0 1357911 more finely tuned pitch perception capacities and Frequency are more accurate at vocal pitch-matching tasks.19 Combined with findings from this study, the data FIGURE 2. LTAS of the talented and nontalented groups when support the views of many singing pedagogues that singing “America the Beautiful.” The numbers on the X-axis represent the successive harmonic peaks, with the fundamental pitch control and timbre are the two most salient frequency corresponding to the first harmonic. aspects of the singing voice for perceptual judgments of talent in untrained singers.8 nontalented singers. We used a relatively small Hollien22 has suggested that superior singing abil- sample size for group comparisons, and as such any ities are a product of both talent and training. In the generalizations based on the results should be made literature of voice science where physiological and with caution. As expected, a significant difference perceptual variables related to the singing voice are was found between the two groups. The talented investigated, persons with and without voice training subjects were collectively different from the nontal- are often compared. These subgroups are usually ented subjects with regard to how higher spectral labeled as trained and untrained singers, singers and partials were tuned. There was greater emphasis in nonsingers, or professional singers and nonsingers. the 2–4-kHz harmonic band, which suggests more Within a population of untrained singers, some will “ring” to the singing voice. Greater energy in this express singing talent, whereas others will not. spectral region has been argued to reflect on singing Although numerous studies have found differences voice quality.13 IftheSPRisameasureofsinging between persons with and without training, few stud- voice quality, it may be possible to infer that the ies have separated untrained subjects into those who energy relationship between the lower and the higher have and do not have expressed singing talent. In harmonics was a perceptually meaningful factor to addition, although there is great public interest in the judges who placed each untrained person into singing talent, little objective empirical evidence their respective group. Specifically, those with sheds light on the factors responsible for ex- greater energy in the higher harmonics (correspond- pressed singing talent in untrained persons. This ing to a lower spectral tilt) were judged as expressing study and previous studies in this line of investiga- singing talent. tion have presented data that support an argument The results of this study are consistent with previ- for more research in this topic, as numerous physio- ous investigations that have studied harmonic tuning logical and perceptual differences may exist, yet they measures (eg, the formant of the singer) in trained are poorly defined, between untrained talented and and untrained singers. These studies have typically nontalented singers. reported greater energy in the region of 2–4 kHz in To date, the line of investigation pertaining to trained voices.9,10,14 The results of this study are this study has found several interesting results. It also in line with a body of research that has demon- seems that professional teachers of singing rely more strated that trained singers, in comparison with on pitch control and voice quality than on other nonsingers, exhibit different physiological capabili- factors when making judgments of singing talent.8 ties during the act of singing.14–21 Measures of har- When testing actual abilities of talented and nontal- monic tuning, such as SPR, offer an indirect method ented persons, talented singers are more accurate

Journal of Voice, Vol. 20, No. 1, 2006 CHRISTOPHER WATTS ET AL88 than nontalented singers when matching pitch, 7. Yarbrough C, Green G, Benson W, Bowers J. Inaccurate which is likely related to greater pitch control abili- singers: an exploratory study of variables affecting pitch- ties, including perceptual acuity.8,19 The results of this matching. Bull Council Res Music Educ. 1991;105:23–33. 8. Watts C, Barnes-Burroughs K, Andrianopolous M, Carr M. study also delineate differences between these two Potential factors related to untrained singing talent: a survey groups during the act of singing. Specifically, tal- of singing pedagogues. J Voice. 2003;17:298–307. ented singers seem to produce lower spectral tilt 9. Lundy D, Roy S, Casiano R, Xue J, Evans J. Acoustic when resonating the voice source, which most likely analysis of the singing and speaking voice in singing stu- contributes to the perception of “ring” in the voice dents. J Voice. 2000;14:490–493. 10. Omori K, Ashutosh K, Carroll L, Riley W, Blaugrund S. or, at a minimum, a more pleasing vocal timbre, and Singing power ratio: quantitative evaluation of singing possibly increases the likelihood for being per- voice quality. J Voice. 1996;10:228–235. ceived as expressing singing talent. 11. Lofqvist A, Mandersson B. Long-time average spectrum Although this study involved a small sample size, of speech and voice analysis. Folia Phoniatr. 1987;39: the significant findings point to the need for more in- 221–229. depth and comprehensive studies that investigate 12. White P. Long-term average spectrum analysis of sex- and gender-related differences in children’s voices. Logoped physiological and perceptual differences in talented Phoniatr Vocol. 2001;26:97–101. and nontalented persons to understand why some, 13. Ackerlund L, Gramming P. Average loudness level, mean without training, can use their vocal mechanism to fundamental frequency, and subglottal pressure: compari- produce a talented singing voice and others are not. son between female singers and nonsingers. J Voice. 1994; This knowledge is important for the area of voice 8:263–270. 14. Brown W, Rothman H, Sapienza C. Perceptual and acoustic science, as it will add to our basic understanding of study of professionally trained versus untrained voices. the voice organ and the physiological bases that are J Voice. 2000;14:301–309. most crucial for controlling the voice during the act 15. Brown W, Hunt E, Willimans W. Physiological differences of singing. between the trained and untrained speaking and singing voices. J Voice. 1988;2:102–110. 16. Brown W, Morris R, Hicks D, Howell E. Phonational pro- REFERENCES files of female professional singers and nonsingers. J Voice. 1993;7:219–226. 1. Watts C, Murphy J, Barnes-Burroughs K. Pitch matching 17. Sulter A, Schutte H, Miller D. Differences in phonetogram accuracy of trained singers, untrained subjects with talented features between male and female subjects with and without singing voices, and untrained subjects with nontalented vocal training. J Voice. 1995;9:363–377. singing voices in conditions of varying feedback. J Voice. 18. Sundberg J. Maximum speed of pitch changes in singers 2003;17:185–194. and untrained subjects. J Phonet. 1979;7:71–79. 2. Goetze M, Cooper N, Brown C. Recent research on singing 19. Watts C, Moore R, McCaughren K. The relationship be- in the general music classroom. Bull Council Res Music tween vocal pitch-matching skills and pitch discrimination Educ. 1990;104:16–37. skills in untrained accurate and inaccurate singers. J Voice. 3. Davies A, Roberts E. Poor pitch singing: a survey of its In press. incidence in school children. Psychol Music. 1981;3:24–36. 20. Michel J, Dawes P. Accuracy of pre-phonatory laryngeal 4. Flowers P, Dunne-Sousa D. Pitch-pattern accuracy, tonality, adjustments. In: Lawrence V, ed. Transcripts of the Seventh and vocal range in preschool childrens singing. J Res Music Symposium on Care of the Professional Voice. New York: Educ. 1990;38:102–114. The Voice Foundation; 1978:85–87. 5. Joyner D. The monotone problem. J Res Music Educ. 1969; 21. Murry T. Pitch-matching accuracy in singers and nonsing- 17:115–124. ers. J Voice. 1990;4:317–321. 6. Price H. Interval matching by undergraduate nonmusic 22. Hollien H. That golden voice—talent or training? JVoice. majors. J Res Music Educ. 2000;48:360–372. 1993;3:195–205.

Journal of Voice, Vol. 20, No. 1, 2006