Listener Perception of the Effect of Abdominal Kinematic Directives on Respiratory Behavior in Female Classical Singing

Sally Collyer, Dianna T. Kenny, and Michaele Archer, Sydney, Australia Summary: Breath management training in classical singing is becoming increasingly physiologically focused, despite evidence that directives focusing on chest-wall kinematic (ribcage and abdominal) behavior effect minimal change in acoustical measures of singing. A direct and proportionate relationship between breathing behavior and vocal quality is important in singing training because singing teachers rely primarily on changes in sound quality to assess the efcacy of breath management modication. Pedagogical opinion is also strongly divided over whether the strategy of retarding the reduction in abdominal dimension during singing has a negative effect on vocal quality. This study investigated whether changes in abdominal kinematic strategy were perceptible and whether listeners preferred a particular strategy. Fourteen experienced singing teachers and vocal coaches assessed audio samples of ve female classical singers whose respiratory kinematic patterns during singing had been recorded habitually and under two simple, dichotomous directives: Gradually drawing the abdomen inward and gradually expanding the abdomen, during each phrase. Listeners rated the singers on standard of singing and of breath management. Ratings analysis took into consideration changes in kinematic behavior under each directive determined from the respiratory recordings. Listener ratings for two singers were unaffected by directive. For three singers, ratings were lower when the directive opposed habitual kinematic behavior. The results did not support the pedagogical assumption of a direct and proportional link between respiratory behavior and standard of singing or that the abdomen-outward strategy was deleterious to vocal quality. The ndings demonstrate the importance of considering habitual breathing behavior in both research and pedagogical contexts. Key Words: PerceptionBreathingSingingPedagogy.

INTRODUCTION Breath management training is fundamental to classical singing pedagogy.18 Good breath management is considered essential for vocal health and for optimal vocal quality. Breath management training depends on directives by which teachers convey concepts and modify singers respiratory behavior. Such directives can be highly stylized such as oating the voice like a ball on a jet of water,6(p60) but in recent years directives have become increasingly physiologically orientated with particular focus on chest-wall kinematic (CWK), that is, abdominal and ribcage behavior.1,2,7,9 Yet Swank10 found that CWK directives effected less change in acoustical measures of voice than did more generalized directives. This would seem to question whether the current trend for physiological focus in breathing training for singing results in perceptible vocal changes. The assumption that change in breathing behavior has a direct effect on vocal quality is a cornerstone of singing pedagogy and plays a crucial role in singing training. At a day-to-day level in the singing studio, the primary feedback tool for singing teachers in assessing the effect of any instruction or concept is to monitor changes in the singers vocal
Accepted for publication October 15, 2009. Part of this research work was presented at the Seventh International Congress of Voice Teachers (ICVT7) held 1519 July, 2009 in Paris. From the Sydney Conservatorium of Music, The University of Sydney, Sydney, Australia. Address correspondence and reprint requests to Sally Collyer, PO Box 156, Box Hill, Victoria 3128, Australia. E-mail: sallycollyer@yahoo.com.au Journal of Voice, Vol. 25, No. 1, pp. e15-e24 0892-1997/$36.00 2011 The Voice Foundation doi:10.1016/j.jvoice.2009.10.006

quality. Given the centrality of breathing training in singing, it is important that singing teachers be condent that aural monitoring accurately reects the type and amount of change in the singers breathing behavior. The two, dichotomous approaches to abdominal directive can be described as abdomen inward (abdomen-iN) and abdomen outward (abdomen-ouT). Variations of these directives (colloquially, belly-in and belly-out) have received attention in both pedagogical and singing research literature.1113 They reect the awareness in singing pedagogy of the two-fold contribution of abdominal activity to vocalization. Firstly, contracting the abdominal wall raises the diaphragm, increasing intrathoracic pressure and thus contributing to subglottal pressure for vocalization. Abdominal contraction also lets the singer use lung volumes (LVs) below end-expiratory level. However, teachers are aware that abdominal activity can be negligible and even absent in vocalization at comfortable pitch, loudness, and timbre. Directing the singer to draw the abdomen-iN during each phrase encourages continuous abdominal activity throughout a phrase, including where musical demands might not otherwise maintain engagement. Secondly, the abdominal wall acts as a sophisticated antagonist to the diaphragm and ribcage facilitating subtle, rapid, and precise adjustments in subglottal pressure with changes in vocal demands, LV, and recoil forces. Directing the singer to expand the abdomen during each phrase encourages diaphragmatic coactivation, which has been associated with increased subglottal pressure, increased peak transglottal airow, and increased stability of formant frequencies.14,15 Pedagogically, the abdomen-ouT directive has been heavily criticized for deleterious effects on vocal quality

e16
(eg,7) but it also has strong advocates (eg,4). The lack of perceptual study of the abdomen-ouT directive, in particular, means these criticisms are untested. Swank10(p13) tested the following abdominal-inward directive:
As you sing the next tones, feel the muscles of the abdominal wall contract, with gradual compression, beginning very low and moving inward and upward gradually. Let the muscles low in the abdomen help you lift the airow upward and out.

Journal of Voice, Vol. 25, No. 1, 2011

Swank10 found that this directive was less effective in eliciting vocal change on a range of acoustical measures, including tone duration and energy in the region 28003300 Hz associated with the singers formant.16 Two points arise from this observation. Firstly, the relationship between acoustical measures and listener preferences is highly complex and not yet fully understood.1719 Thus, the acoustical results reported by Swank cannot be assumed to imply a deterioration in vocal quality. For example, an increase in tone duration can occur when a singer increases vocal-fold medial compression, but a voice that sounds pressed would not be considered as an improvement in classical singing. This means that the perceptual implications of the acoustical changes observed by Swank are unclear. Secondly, Swank10 did not measure respiratory behavior. In an earlier study,20 we investigated the effects on female classical singers CWK behavior of abdomen-iN and abdomen-ouT directives. Our abdomen-iN directive was comparable with that of Swank10 and asked that the singers steadily pull the abdominal wall inward for each phrase whilst singing, whereas the abdomen-ouT directive asked that the singers steadily expand the abdominal wall during singing. The abdomen-ouT directive was of interest given the conicting opinions of its appropriateness in singing pedagogy, discussed above. We found that highly trained singers can alter aspects of their CWK behavior under directive and can sustain these changes throughout a song. However, the changes that occurred were a function of the interaction of the directive and the singers habitual (ie, undirected) CWK behavior. Also, the amount of change varied among singers, with some singers showing large change and some very little change under each directive. Our results showed that assessment of the inuence of any directive on vocal quality must take into account how and to what extent CWK behavior has actually changed, especially given the wide differences in singers kinematic strategies that have been observed.2123 It seems reasonable to assume that the singers studied by Swank10 would have shown a similar interactive effect of habit and directive, but without CWK data the extent to which this might have affected the acoustical results cannot be determined. The relationship between CWK behavior under a physiologically based directive and perception of voice does not appear to have been investigated. Abdominal directives have been studied for inhalation, with mixed outcomes in terms of effect on vocal quality. Thomasson24 found no signicant effect for professional singers on a range of voice source parameters, including vertical laryngeal position. Iwarsson,25 however, found a higher

vertical laryngeal position for speakers in the abdomen-ouT condition (after controlling for LV), which post hoc investigation associated with changes in posture. Again, neither of the studies incorporated an audio perceptual component. Studies of nonphysiologically based directives have found perceptible changes in vocal quality with changes in respiratory behavior. Foulds-Elliott et al26 found that instructing professional singers to perform with emotional connection led to higher LV at the start of phrases and lower at the end and consequently larger expenditure of air than when merely singing loudly. They also found that experienced listeners could distinguish audio samples with and without emotional connection, although interestingly the perceptual cues reported by the listeners did not include any assessment of changes in breathing during phonation. Another nonphysiological directive often used in pedagogy is to support the voice with the breath, considered essential for the singing voice13 despite the elusiveness of an agreed definition for the term support. Thorpe et al asked singers to imagine they were projecting their voices over a large or small orchestra,27(p100) a directive which the authors associated with greater and lesser abdominal support, respectively. Kinematic patterns were similar in both conditions, but greater projection was associated with higher LV at the end of phrases and concomitantly smaller expenditure of air. Despite the association of projection with abdominal activity, they found LV changes to be mostly because of changes in ribcage dimension. Even so, changes in sound level and frequency spectrum characteristics could not be correlated with changes in respiratory parameters. Grifn et al28 (p51) found that the direction to sing with support elicited no changes in respiratory behavior, despite noting that the subjects in this study clearly believe that a supported singing voice is . produced by managing breathing activity differently. They did, however, nd signicant changes in acoustic and laryngeal measures (including peak glottal airow) that implied changes in voice quality. Although they did not conduct a perceptual study, strong listener agreement in assessing the presence and degree of support present in audio samples has been reported by Sand and Sundberg.29 Sonninen et al30 also found laryngeal and acoustic differences in singers asked to sing with and without support, but listener perception was more complex. Although listeners were generally able to identify supported and unsupported singing samples, their assessment of the degree of support (0100%) present in a sample varied widely. Indeed, three of the seven singers could not differentiate between supported and unsupported singing when listening to their own recordings 2 months after data collection. The authors noted that [t]hese samples also caused difculties to other listeners.30(p231) Respiratory behavior was not measured, so the extent of change in the singers breathing under directive is unknown. Interestingly, they found no distinction between ratings for voice quality and for support, suggesting that the terms and concepts were interchangeable. In summary, studies of nonphysiologically based directives have produced mixed results in terms of changes in respiratory

Sally Collyer, et al

Listener Perception of Abdominal Directives in Singing

e17

behavior and consequent changes in vocal quality. The difculty with nonphysiologically based directives from a singing training perspective is that they lack clear and agreed denition. This has led singing teachers to rely on more direct, physiologically based instruction, but the perceptible effects of such instructions are unknown. This study asked classical singing teachers and vocal coaches to assess the standard of singing and of breath management of singers performing under three conditions: habitually (ie, undirected) and under two simple, specic, and dichotomous abdominal kinematic directives: abdomen-iN and abdomenouT. The audio samples were derived from the companion study20 that analyzed the type and extent of change in respiratory behavior elicited by the directives. That study found that the singers habitual abdominal behavior lay on a continuum between the extremes elicited by the two directives, generally resembling one directive more than the other. Therefore, it was hypothesized that listeners would be able to detect a difference in standard of singing and standard of breath management proportionate to the degree of CWK change exhibited by the singer. It was further hypothesized that listeners would rate habitual singing higher than nonhabitual singing and, in turn, would rate singing under the abdomen-iN directive higher than under the abdomen-ouT directive. METHOD The study was approved by the Human Ethics Committee of The University of Sydney. Fourteen singing teachers and vocal coaches experienced with classical singing voice assessed audio samples of ve female classical singers under three breathing conditions. All participants received an information sheet and gave written informed consent. Audio samples Audio and respiratory recordings were made of ve professional female classical singers: three sopranos (Singers A, B, and E), one mezzo-soprano (D), and one contralto (C), ranging from 31 to 44 years of age. All singers held postgraduate qualications in singing and had between 7 and 20 years of professional performing experience. At the time of recording, Singer D was in her 19th week of pregnancy but was maintaining her full performance schedule. Recordings were made at the audiorecording laboratory at the Sydney Conservatorium of Music (The University of Sydney), measuring 4.84 3 4.55 3 2.64 m and hung with sound-absorptive curtains (http://www.jands. com.au/). Respiratory measurements were made using respiratory inductance plethysmography (Inductotrace; Ambulatory monitoring system Inc., Ardsley, NY). Audio recordings were made with a pair of cardioid microphones (Neumann, SKM 140, Georg Neumann GmbH, Berlin) in ORTF (Ofce de le vision Franc Radiodiffusion Te aise) conguration31 placed 2.75 m from the singer. Sound intensity has been found to inuence listener assessment of voice.32 Analysis of the audio recordings20 conrmed that the equivalent sound level calculated from long-term average spectra of the last four phrases was consistent across breathing condition (F2,8 2.795,

P 0.12), that is, any listener preferences in this study could not be attributed to differences in sound level between performances. Each singer sang two takes of Caccinis Ave Maria in g minor (unornamented, no repeat) without instruction (Habitual) and two takes under each of the following directives: N Abdomen-iN: steadily pull the abdominal wall inward during phonation, using an additional inward pull if required for additional support and T Abdomen-ouT: steadily expand the abdominal wall during phonation, using an additional outward expansion if required for additional support. CWK plots of ribcage and abdominal dimensions during the last four phrases of the song were made for each singer.20 Examples are given in Figure 1. CWK plots show the contribution of abdominal (abscissa) and ribcage (ordinate) dimensional change to change in LV, with the axes scaled to contribution capacity (after22). Kinematic traces that run parallel to the 45 isovolume lines represent a change in ribcage:abdomen conguration but no change in LV. Traces that move to the left of an isovolume line represent a net decrease in LV even where the ribcage or abdominal dimension might be increasing. The small cross represents ribcage:abdomen conguration at end-expiratory level, that is, at the end of expiration in quiet breathing. The plots were used to select the three takes (one per breathing condition) that showed the greatest variance in kinematic behavior, and the audio samples from these takes were used for the perceptual study. That is to say, the audio samples were chosen according to respiratory behavior not according to the sound. Two audio CDs were constructed using the last four phrases of the chosen take under each breathing condition (H, N, and T). The order of singers was different on each CD, and the order of breathing condition was randomized within that. Audio samples of the rst two singers on each CD (Singers A and B on CD#1 and E and D on CD#2) were repeated. Odd-numbered listeners heard CD#1 (Singers A, B, C, A, D, E, B) and even-numbered listeners heard CD#2 (Singers E, D, C, B, E, A, D). Repeating two singers samples allowed us to assess how consistently listeners assessed the same sample on different hearings. Randomizing and using two CD sets minimized the possibility that results reected sample order rather than inherent vocal quality. Furthermore, the three individual samples for each singer were preceded by a composite track that concatenated the three samples (with a 4-s pause between) in the same order. Thus, each CD contained 28 tracks: seven singers 3 four tracks. Studies have found that listeners tend to use the rst sample heard as a context for subsequent samples17,33 and that listeners internal standards of voice quality are inherently unstable.34 Playing the concatenated track rst was designed to have all three samples set the comparative context and to provide a wash-out of the inuence of assessment of a previous singer on a subsequent singer.35 Listener data The 14 listeners consisted of 11 classical singing teachers, of whom six maintained a national and/or international

e18
Singer A Singer D

Journal of Voice, Vol. 25, No. 1, 2011

Singer B

Singer E

25 50 75

Decreasing LV

100% LV

0% LV

25 50 75 % Abdominal capacity
Singer C

FIGURE 1. Examples of the changes in CWK behavior elicited by

abdominal kinematic directive for the ve singers. (Thick solid line habitual, broken line abdomen-iN, ne solid line abdomen-ouT; LV, lung volume.) Construction of the plots is described in Method: Audio samples. These plots compare the second-last phrase of the rst or second take (whichever was used in the perceptual study; refer to Method: Audio samples) under the three conditions, showing the nature and extent of kinematic change elicited by directive and underlying the audio samples. Axes are scaled to percentage of abdominal and ribcage capacity.14

performance schedule, two chorus-masters, one accompanist, and one singing-voice researcher specializing in classical voice. All but one held tertiary qualications in music, the exception being a singing teacher who had retired from performing after a career, including leading roles at Covent Garden and the Metropolitan Opera, and all were employed full-time professionally in their respective elds. Hearing was not tested, but all reported normal hearing and all rely on accurate hearing for their professions. Data collection was conducted in quiet rooms with only the participant and one researcher present, using a CD player (Sony Model CFD V8; Sony CFD S300; or Sony CD Walkman DEJ885 W, Sony Corp., Tokyo) and headphones (Sennheiser, HD 650, Sennheiser Electronics GmbH, Wedemark-Wennebostel, Germany). Listeners rst heard the composite track containing all three samples in the same order in which they would be presented individually. They then heard the rst sample that they assessed using ve visual analog (VA) scales with gradations numbered 010 from left to right. The greater resolution of continuous scales has been found to improve listener agreement over equal-interval scales by not

forcing a choice between ratings on samples perceived to fall between intervals.3537 Listeners were asked to rate the rst sample by placing a vertical mark and the number 1 above it on each scale, before moving to the second sample. During explanation of the task, a dummy completed sheet was used to illustrate how to complete the form. Sample assessments on this dummy form included a mix of widely and closely spaced marks, so as to encourage but not to prescribe that listeners make full use of a scales range. Listeners were not constrained from marking between gradations (eg, at a location equating to 6.3) or from giving two samples the same score. The rst scale asked, Please assess the standard of singing for each sample and the second asked, Please assess the standard of breath management for each sample. (The three remaining scales addressed tone color, vibrato rate, and vibrato extent and are not included in this study.) Standard of singing was positioned rst as Stanley et al38 found judges make an overall assessment before attending to specic criteria. Scales also had the cues poor (underneath 0), average (under 5), and excellent (under 10). Listeners were given as much time as they wished to assess each sample before hearing the next sample, but tracks were not repeated. Fresh sheets were provided for each set of three samples, and listeners could not refer to previous sheets. Data collection generally took 11.5 hours. The listener criteria standard of singing and breath management were deliberately general in nature, and no clarication or further explanation of the terms was provided. Listeners were invited, but not required, to add comments in the space provided beneath each scale if they wished to clarify aspects of their assessments. Intra- and interjudge reliability in assessing singers have been found to be higher for general (overall) assessments than for specic criteria in singing39 and in speech pathology.17 Listeners have also been found to differ as to whether they assess on performance or intrinsic vocal quality,39 and it was intended that the term standard of singing should encompass both aspects. The implicit inclusion in listeners assessments of factors not relevant to this study, such as diction and phrasing, was not considered signicant because comparisons were between performances by the same singer. Listeners were made aware of the purpose of the study and of the breathing directives that had been given to the singers. This ensured that listeners worked from an equal knowledge base and encouraged them to focus on the breathing behavior of the singers. They were informed that respiratory measurements had been made of the singers but not whether the directives had elicited changes in respiratory behavior. Because only ve singers were used, it was anticipated that experienced listeners would be likely to recognize a repetition. Therefore, listeners were explicitly told that two singers samples would be heard twice, again to ensure that all listeners worked from an equal knowledge base. Data analysis The score for each listener criterion for each sample was measured by ruler in millimeters and input into a customized spreadsheet (Excel 2003, Microsoft) that converted the result

% Ribcage capacity

Sally Collyer, et al

Listener Perception of Abdominal Directives in Singing

e19

into a percentage. This was checked against a visually estimated score. Comments were also recorded on the spreadsheet. Statistical analysis used linear mixed modeling in SPSS (v14.0 for Windows) (SPSS Inc., Chicago, IL) with repeated measures. Model testing for best-t of covariance structure used the corrected Akaike Information Criterion40 because of the small sample size. A criterion of P < 0.05 for pairwise comparisons was adjusted (Bonferroni) for three multiple comparisons (H vs T, N vs T, and H vs N), giving a signicance criterion of P < 0.017. Intralistener (test-retest) repeatability was assessed by correlation (Pearsons r) and by modeling with singer and breathing condition as xed factors and hearing (rst and second) as the repeated measure. (Only scores from the second hearing were used in the other analyses.) The relationship between the two listener criteria was similarly tested by correlation and by modeling, with the score for each criterion as the repeated measure. Interlistener consistency for each listener criterion was assessed from covariance parameters obtained by a mixed effects model, with singer and listener as random factors and breathing condition as a xed factor.41 Listener scores were analyzed using linear mixed modeling with xed factors of singer and breathing condition. Data were arranged in the order T-N-H to reect assumptions that the abdomen-ouT behavior is the least-favored pedagogically and that habitual behavior would score higher than the nonhabitual behaviors. For both listener criteria, the scaled identity covariance structure was the best t for both random and repeated effects and singer (but not breathing condition) was a random factor. Signicant interactions were assessed by pairwise comparisons. RESULTS CWK patterns Examples of the CWK patterns of the singers from the audio samples used in this study are plotted in Figure 1. The plots show the second-last phrase drawn from the three takes (one per breathing condition) used for the perceptual study. Respiratory analysis reported in the companion study20 found that the directives led the singers to alter the abdominal and ribcage dimensions at the start but not at the end of phrases and that LV measurements were unchanged. Figure 1 illustrates how habitual kinematic strategy changed with directive for each of the singers, highlighting similarity and difference between habitual and directive patterns. This information allowed the perceptual results to take into consideration how much behavioral change was actually elicited. Singer As habitual pattern lay very close to the abdomen-iN end of the continuum; accordingly, the abdomen-ouT directive had a major effect on her kinematic behavior, completely removing ribcage paradoxing and disrupting smooth coordination between abdomen and ribcage. Singer B showed clear differences between the three conditions: abdominal paradoxing (expansion during phonation) under N suggested that she found the sharp initial decrease unsustainable but there was no such disruption under T, although it is clear that she was resisting her habitual tendency for abdominal contraction. Singer C showed

the most striking differences between conditions, with ribcage paradoxing under N and abdominal paradoxing under T. By contrast, Singer D showed the least change across condition, although adjustments that were characteristic of other singers (increased ribcage paradoxing in N, delayed abdominal contraction in T) were clear in her kinematic plots. Lastly, Singer Es kinematic trace for N generally paralleled her habitual trace. In summary, habitual patterns seemed to lie toward the abdomen-iN end of the continuum for Singers A, D, and E, toward the abdomen-ouT end for Singer B, and midway for Singer C. Intra- and interlistener consistency For both listener criteria, test-retest correlations were moderate (standard of singing r 0.683, P < 0.001; breath management r 0.674, P < 0.001), and there were signicant differences between scores for the rst and second hearing. Model testing resulted in a scaled identity covariance structure, with singer (but not breathing condition) as a random factor. For standard of singing, main effects of singer (P < 0.001), breathing condition (P 0.016), and hearing (P < 0.001) were all signicant, as were interactions of singer3condition (P < 0.001) and singer3hearing (P 0.034). For breath management, singer was the only signicant main effect (P < 0.001), but interactions of singer3condition (P 0.003) and singer3hearing (P 0.008) were again signicant. The interaction of singer3condition3hearing was not signicant for either criterion and was removed from the nal models. Pairwise comparisons for the singer3hearing interaction found that listeners rated standard of singing higher on second hearing for two of the four singers (Singer D mean difference 12.2, standard error [SE] 2.9, P < 0.001; Singer E mean difference 6.9, P < 0.001), whereas listeners rated the breath management higher on second hearing for Singer D (mean difference 10.1, SE 3.4, P <0.001). Singers D and E were used as repeated samples for even-numbered listeners, suggesting that the effect might have been because of one or two listeners. However, visual examination of plotted data did not indicate that any of the listeners produced unusual results, and the results were unaltered by excluding each of the even-numbered listeners in turn and rerunning the analysis. Interlistener variance accounted for 17.9% (P 0.026) of the total variance for standard of singing and 19.0% (P 0.029) for breath management. Comparison of listener criteria Listener scores for the two criteria were highly correlated (r 0.857, P < 0.001). However, modeling found a signicant main effect of criterion (P < 0.001) after controlling for singer and condition, and a signicant interaction of criterion3singer (P < 0.001) but not of criterion3condition. These results of high correlation with systematic difference between the criteria can be seen in Figure 2 that plots the estimated marginal means and SE for the two criteria, split by singer and breathing condition. Listeners rated breath management more critically (lower) than standard of singing for all singers except Singer C, regardless of breathing condition. This occurred despite the

e20
100

Journal of Voice, Vol. 25, No. 1, 2011

80

Score (out of 100)

60

40

20

0 T N A H T N B H T N C H T N D H T N E H

Singer (by breathing condition)

FIGURE 2. Estimated marginal means and standard errors of listener scores (out of 100) for standard of singing (solid squares) and standard of
breath management (crosses). Scores are grouped by singer (AE), then by breathing condition (abdomen-ouT, abdomen-iN, and habitual). Singers B and E scored signicantly higher than did Singers A, C, and D on both listener criteria. Scores for the two criteria were highly correlated, but listeners rated breath management more critically (lower) than standard of singing for all singers except Singer C.

two scales being vertically aligned on the page, so that the disparity would have been visible to the listeners.

Criterion scores Kolmogorov-Smirnov tests were nonsignicant (standard of singing Z 0.51; breath management Z 0.39) and visual examination of residual plots identied no trends, conrming normality of residual distribution. Tests of xed effects are set out in Table 1. For both criteria, there was a signicant main effect for singer (P < 0.001). Pairwise comparisons found that the scores for Singers B and E were signicantly higher than for Singers A, C, and D, and this can be seen in Figure 2. There was a signicant main effect of breathing condition for standard of singing (P 0.019) but not for breath management (P 0.102). However, the interaction of singer3breathing condition was signicant in both criteria (P < 0.001), that is, there were signicant differences between condition but only for some singers.

These differences are broken down in the pairwise comparisons for the singer3breathing condition interaction in Table 2 and can also be seen in Figure 2. Pairwise comparisons were tested at a Bonferroni-adjusted signicance of P < 0.17. For standard of singing, Singer A scored signicantly lower in the abdomen-ouT condition than in the other conditions, whereas Singer D scored higher in the habitual condition than under either directive. For breath management, Singer A again scored lower in the abdomen-ouT condition than in the other conditions. Singer D again scored higher in the habitual condition than under the abdomen-iN directive, but scores under the abdomen-ouT directive were not signicantly lower, as they were for standard of singing. Singer B scored higher under the abdomen-ouT directive than under the abdomen-iN. Her mean score under T was also higher than for her habitual samples but failed to reach Bonferroni-adjusted signicance (P 0.17). Singer Es habitual singing was rated more highly than her singing under T for both criteria but again the differences were not statistically signicant after adjusting for multiple comparisons.

TABLE 1. Linear Mixed Modeling of the Two Criteria Found a Signicant Interaction of Breathing Condition and Singer, That Is, There Were Signicant Differences in Ratings on Both Criteria Between Breathing Condition but Only for Some Singers Standard of Singing Fixed effects Breathing condition Singer Condition3singer F2,128.9 4.085, P 0.019 F4,68.6 22.947, P < 0.001 F8,128.9 4.969, P < 0.001 Standard of Breath Management F2,128.6 2.327, P 0.102 F4,70.4 15.645, P < 0.001 F2,128.6 3.425, P 0.001

Sally Collyer, et al

Listener Perception of Abdominal Directives in Singing

e21

TABLE 2. Pairwise Comparisons of the Interaction of Breathing Condition and Singer H vs N Singer Standard of singing A B C D E Mean Difference, P 3.2, P 0.326 5.2, P 0.106 4.2, P 0.194 11.6, P 0.001* 2.9, P 0.375 H vs T Mean Difference, P 10.6, P 0.001* 2.5, P 0.437 4.2, P 0.200 9.3, P 0.005* 7.2, P 0.027** 12.1, P 0.003* 6.5, P 0.100 3.7, P 0.349 6.7, P 0.093 8.5, P 0.033** N vs T Mean Difference, P 13.8, P < 0.001* 7.8, P 0.017** 0.1, P 0.988 2.3, P 0.482 4.3, P 0.180 10.1, P 0.011* 10.0, P 0.012* 0.1, P 0.981 4.9, P 0.219 5.9, P 0.138

Standard of breath management A 1.9, P 0.629 B 3.5, P 0.381 C 3.6, P 0.362 D 11.6, P 0.004* E 2.6, P 0.509

*Signicant at P < 0.17; **Signicant without Bonferroni adjustment for multiple comparisons.

DISCUSSION Breath management is acknowledged as fundamental to singing, but its training relies on the assumption that a singers breathing behavior has a direct and proportionate effect on the singers vocal quality. The corollary, that changes elicited by physiologically based CWK directives can be monitored immediately by the singing teachers assessment of vocal quality, is especially important because breathing instruction is becoming increasingly dened in terms of physiology. This study tested whether the type and degree of change in CWK pattern elicited by two simple, specic, and dichotomous abdominal kinematic directives proportionately matched the assessment of standard of singing and of breath management of female professional classical singers, assessed from audio samples by listeners experienced with the classical singing voice. The singers sang under three breathing conditions: without instruction (habitually), pulling the abdominal wall inward during phonation (abdomen-iN), and expanding the abdominal wall during phonation (abdomen-ouT). The rst hypothesis was that listeners would be able to detect a difference in standard of singing and standard of breath management proportionate to the degree of CWK change exhibited by the singer. Key to this was to take into consideration the degree of similarity between each directive and the singers habitual CWK strategy. The results for Singers A and B were consistent with the hypothesis. Singer As habitual pattern was very similar to her strategy under the abdomen-iN directive and both were markedly different from the abdomen-ouT directive. Likewise, listener ratings found no difference between her singing in H and N but found both differed from her singing under T. Singer Bs CWK pattern differed less between conditions than did Singer As, but her habitual pattern tended to be closer to T. Likewise, listeners rated her breath management to be better in T than N but not different under directive than habitually. Although ratings for the standard of singing showed the same

trend, the smaller differences were not statistically signicant after Bonferroni adjustment for multiple comparisons. Strictly speaking, Singer Es results could also be considered as consistent with the hypothesis. Her habitual pattern more closely resembled that under the abdomen-iN directive, and listener scores averaged highest for her habitual singing and lowest for her abdomen-ouT singing. However, the differences between CWK patterns were small and between listeners scores were not signicantly different after Bonferroni adjustment. Her results, therefore, do not contradict but also do not support the hypothesis. The results for Singers C and D were not consistent with the hypothesis. Singer C showed the most change in CWK strategy under directive, yet there was no signicant difference in listener ratings. Singer D showed the least change in CWK pattern, yet listeners found the standard of both singing and breath management in the nonhabitual conditions to be inferior to habitual behavior. Thus, the results for Singers A, B, and E appeared consistent with the hypothesis but the results for Singers C and D were not. It was also hypothesized that listeners would rate habitual singing higher than nonhabitual singing and, in turn, would rate singing under the abdomen-iN directive higher than under the abdomen-ouT directive, following pedagogical opinion that abdomen-ouT behavior is deleterious to vocal quality. Again, the results were mixed. In terms of habitual versus nonhabitual breathing, only Singer D scored signicantly higher for H than under either directive. However, the strong similarity between Singer As H and N CWK strategies meant that her signicantly lower scores under T also support this hypothesis. Lastly, Singer Es results showed higher listener preference for habitual than nonhabitual breathing but the differences were not statistically signicant. Mean ratings for Singers B and C went against the hypothesis but again the differences were not statistically signicant.

e22
In terms of ratings under abdomen-iN being higher than under abdomen-ouT, only Singer As ratings supported this and only at face value. That is to say, the strong similarity between her H and N CWK strategies noted above suggests that the lower listener rating associated with the abdomen-ouT directive might occur with any nonhabitual directive, as was the case with Singer D. Singer Bs results did not support the hypothesis, showing a signicantly higher listener preference for T than N. Again, habitual CWK strategy needs to be taken into consideration, as Singer Bs H pattern was much closer to T than to N. Listeners also preferred T to N for Singer D but not signicantly so. In summary, we found that the ability of experienced listeners to detect changes in breathing behavior depended on the individual singer and on the extent to which the directive deviated from the singers habitual pattern. Furthermore, we found no support for the pedagogical assertion that the abdomen-ouT directive necessarily results in a perceptibly lower standard of singing. Rather, our ndings suggested that the standard of singing was compromised by any directive that elicited CWK behavior antithetical to habitual behavior. Our results raise four issues with respect to the study of breathing training methods in singing: habitual kinematic behavior, limitations of body-surface measurement, participant pool, and improving reliability in perceptual studies in singing.

Journal of Voice, Vol. 25, No. 1, 2011

Participant pool It could be argued that the highly trained singers in our study have already optimized their habitual abdominal behavior. If so, mean scores for breath management for Singers A, C, and D (<50%) imply that there is great room for improvement in their breath management but not through direct modication of abdominal behavior, as is popular in current pedagogical trends. Singers CWK strategies vary widely, and the factors that determine habitual and optimal strategies are not understood.20 Also, the singers in this study were highly trained, and any acoustic changes might lie beneath the threshold of detection.35,37 It is quite possible that the abdominal directives would have a larger and more perceptible effect on a sample of less experienced or beginning students. This does not alter the fact that singers receive ongoing training throughout their professional lives, that breathing training plays a signicant role in this, that teachers rely primarily on sound to guide training,2,7 and that this study did not support the pedagogical assumption that CWK behavior can be reliably inferred from voice perception. Improving reliability in perceptual studies in singing A major concern in the conduct of perceptual studies in voice is listener consistency.17,34,36,37 Intralistener repeatability was unexpectedly high, and interlistener variance was unexpectedly low, given the extensive randomization of the data, the differences in scores between singers, the high resolution of the VA scale, and the high variability reported in perceptual studies of speech36 and singing.30 There was an unexplained intralistener difference, where Singer D received signicantly lower ratings on rst hearing than on second hearing for both listener criteria, although mean differences (12.2 and 10.1, respectively) were less than the equivalent resolution of a 7-point scale.36 The internal standards against which expert listeners compare an audio sample have been found to drift during the rating session as the listening context changes,34 and Kreiman et al37 identied context (in terms of comparison stimuli) as the primary determinant of interlistener agreement. It would seem that presenting all three samples before individual assessment assisted in overcoming the precedence effect of either the rst sample assessed or of the previous singer, without adversely affecting the internal standard by which listeners differentiated the general standard of each singer.34 Nevertheless, more work is needed to understand the inuences on listener assessment of singing voice, not only for research applications but also in adjudication, examination, and audition.36 Consistency was also likely to have been enhanced by the use of a general term, standard of singing.17,39 Sonninen et al30 found that the terms vocal quality and supported voice were highly correlated but they did not test agreement; that is, a score for vocal quality received a comparable but not necessarily equal score for supported voice. We found that listener assessments of standard of singing and breath management were highly correlated, but breath management received systematically lower ratings, that is, was judged more critically. This could imply that, because breath management is only one factor in good singing, the scale for breath

Habitual kinematic behavior Any assessment of a breathing directive should consider CWK strategy throughout the entire phrase, not just excursion-only or LV measures, and should incorporate in the assessment the type and extent of change in CWK strategy elicited by the directive in the context of the singers habitual CWK strategy. By referring to habitual behavior, the apparently contradictory results for Singers A, B, and D were in fact consistent in terms of any directive eliciting nonhabitual behavior.

Limitations of body-surface measurement It must be remembered that body-surface measurement has limitations in describing underlying change in respiratory behavior. This was clearly a factor in the results for Singer D, whose CWK strategies showed small differences between breathing conditions, yet were quite audible to experienced listeners. It is possible that her pregnancy reduced the range of her abdominal movement that could be detected by body-surface measurement, although we are not aware of any studies into the effect of pregnancy on CWK behavior in singers, and the singer was maintaining a full performance schedule. Technological limitations mean that the relationship between CWK strategy, muscular activity, and respiratory proprioception is poorly understood. Likewise, from the teachers perspective, the relationship between CWK strategy dened in terms of contribution to LV change (as used in respiratory research) and in terms of dimensional change (as observed in the teaching studio) needs much clarication before results of CWK studies can be applied in the studio.

Sally Collyer, et al

Listener Perception of Abdominal Directives in Singing

e23

management is in effect a subset of the scale for standard of singing. A subscale would suggest a wider range of scores for breath management. It would also suggest that a given increase in rating for breath management should lead to a smaller increase in standard of singing because it represents only a portion of potential improvement. Although our results (Figure 2) did not show these trends, our study was not designed to investigate these issues. Another possibility is that listeners made an overall assessment that they then justied, consciously or unconsciously. The greater consistency of general criteria as used in our study comes at the cost of identifying and ranking the vocal characteristics that are important to listeners. Our results suggest that teasing out the components of vocal assessment might be complicated by an initial general assessment (solicited or unsolicited) providing a contextual inuence similar to the effect on comparing audio samples discussed above. CONCLUSION In summary, our results question pedagogical assumptions about direct relationships between breathing behavior, breathing training directives, and perceptible changes in vocal quality. There is little examination in research or pedagogy of the circumstances under which training approaches fail or are less successful, despite the value of such information to understanding and rening methods and practices in any domain. Although the enormous variability in habitual CWK behavior in singers has been well documented in research studies, it is yet to be integrated into pedagogical thinking. Our study begins the important work of identifying why certain directives work for some singers and not for others, so that training can be tailored and streamlined. Acknowledgments The authors are grateful to the participants and to Dr Helen Mitchell for her advice on data collection. This study was supported by an Australian Research Council Discovery Grant (DP066559) to Professor Dianna T. Kenny and Dr C. William Thorpe. REFERENCES
1. Blades-Zeller EL. A Spectrum of Voices: Prominent American Voice Teachers Discuss the Teaching of Singing. Lanham, MD: Scarecrow; 2003. 2. Chapman JL. Singing and Teaching Singing: A Holistic Approach to Classical Voice. San Diego, CA: Plural Publishing; 2006. 3. Duey PA. Bel Canto in Its Golden Age: A Study of Its Teaching Precepts. New York, NY: Da Capo; 1980. 4. Greene A. The New Voice: How to Sing and Speak Properly. Milwaukee, WI: Hal Leonard; 1985. 5. Hemsley T. Singing and Imagination: A Human Approach to a Great Musical Tradition. Oxford, UK: Oxford University Press; 1998. 6. Hines J. Great Singers on Great Singing. New York, NY: Limelight Editions; 1982. 7. Miller R. The Structure of Singing: System and Art in Vocal Technique. New York, NY: Schirmer; 1996. 8. Monahan BJ. The Art of Singing: A Compendium of Thoughts on Singing Published Between 1777 and 1927. Metuchen, NJ: Scarecrow; 1978. 9. Spillane KW. Breath support directives used by singing teachers: a Delphi study. NATS J. 1989;45:921. 57.

10. Swank H. Some verbal directives regarding support concepts and their effects upon resultant sung tone. NATS J. 1984;40:1218. 11. Miller R. Solutions for Singers: Tools for Performers and Teachers. Breath management. New York, NY: Oxford University Press; 2004. 14. 12. Hixon TJ, Hoffman C. Chest wall shape in singing. In: Lawrence V, ed. Transcripts of the Seventh Symposium Care of the Professional Voice, 1978 June. New York, NY: The Voice Foundation; 1979:910. Part 1. 13. Hixon TJ. Respiratory Function in Singing: A Primer for Singers and Singing Teachers. Tucson, AZ: Redington Brown; 2006. 14. Leanderson R, Sundberg J, Von Euler C. Role of diaphragmatic activity during singing: a study of transdiaphragmatic pressures. J Appl Physiol. 1987;62:259270. 15. Sundberg J, Leanderson R, Von Euler C. Voice source effects of diaphragmatic activity in singing. J Appl Physiol. 1986;14:351357. 16. Sundberg J. The Science of the Singing Voice. DeKalb, IL: Northern Illinois University Press; 1987. 17. Kreiman J, Gerratt BR, Precoda K, Berke GS. Individual differences in voice quality perception. J Speech Hear Res. 1992;35:512520. 18. Ekholm E, Papagiannis GC, Chagnon FP. Relating objective measurements to expert evaluation of voice quality in western classical singing: critical perceptual parameters. J Voice. 1998;12:182196. 19. Mitchell HF, Kenny DT. The effects of open throat technique on long term average spectra (LTAS) of female classical voices. Logoped Phoniatr Vocol. 2004;29:99118. 20. Collyer S, Kenny DT, Archer M. The effect of abdominal kinematic directives on respiratory behavior in female classical singing. Logoped Phoniatr Vocol. 2009;34:100110. 21. Collyer S, Thorpe CW, Callaghan J, Davis PJ. The inuence of fundamental frequency and sound pressure level range on breathing patterns in female classical singing. J Speech Lang Hear Res. 2008;51:612628. 22. Watson PJ, Hixon TJ. Respiratory kinematics in classical (opera) singers. J Speech Hear Res. 1985;28:104122. 23. Watson PJ, Hixon TJ, Stathopoulos ET, Sullivan DR. Respiratory kinematics in female classical singers. J Voice. 1990;4:120128. 24. Thomasson M. Belly-in or belly-out? Effects of inhalatory behavior and lung volume on voice function in male opera singers. Speech Transm Lab Q Status Prog Rep. 2003;45:6174 [Internet]. Available at: http://www. speech.kth.se/prod/publications/les/qpsr/2003/2003_45_1_061-074.pdf, 2003. Accessed May 4, 2009. 25. Iwarsson J. Effects of inhalatory abdominal wall movement on vertical laryngeal position during phonation. J Voice. 2001;15:384394. 26. Foulds-Elliott SD, Thorpe CW, Cala SJ, Davis PJ. Respiratory function in operatic singing: effects of emotional connection. Logoped Phoniatr Vocol. 2000;25:151168. 27. Thorpe CW, Cala SJ, Chapman J, Davis PJ. Patterns of breath support in projection of the singing voice. J Voice. 2001;15:86104. 28. Grifn B, Woo P, Colton R, Casper J, Brewer D. Physiological characteristics of the supported singing voice: a preliminary study. J Voice. 1995;9:45 56. 29. Sand S, Sundberg J. Reliability of the term support in singing. Logoped Phoniatr Vocol. 2005;30:5154. 30. Sonninen A, Laukkanen A-M, Karma K, Hurme P. Evaluation of support in singing. J Voice. 2005;19:223237. 31. Farina A, Lamberto T. Measurements and reproduction of spatial sound characteristics of auditoria. Acoust Sci Tech. 2005;26:193199. 32. Kempster GB, Kistler DJ, Hillenbrand J. Multidimensional scaling analysis of dysphonia in two speaker groups. J Speech Hear Res. 1991;34:534543. 33. Repp B, Crowder RG. Stimulus order effects in vowel discrimination. J Acoust Soc Am. 1990;88:20802090. anzas-Barroso N, Berke GS. Comparing in34. Gerratt BR, Kreiman J, Anton ternal and external standards in voice quality judgments. J Speech Hear Res. 1993;36:1420. 35. Shrivastav R, Sapienza CM, Nandur V. Application of psychometric theory to the measurement of voice quality using rating scales. J Speech Lang Hear Res. 2005;48:323335. 36. Kreiman J, Gerratt BR, Kempster GB, Erman A, Berke GS. Perceptual evaluation of voice quality: review, tutorial, and a framework for future research. J Speech Hear Res. 1993;36:2140.

e24
37. Kreiman J, Gerratt BR, Ito M. When and why listeners disagree in voice quality assessment tasks. J Acoust Soc Am. 2007;122:23542364. 38. Stanley M, Brooker R, Gilbert R. Examiner perceptions of using criteria in music performance assessment. Res Stud Music Educ. 2002;18:4656. 39. Wapnick J, Ekholm E. Expert consensus in solo voice performance evaluation. J Voice. 1997;11:429436.

Journal of Voice, Vol. 25, No. 1, 2011

40. Clifford MH, Tsai CL. Regression and time series model selection in small samples. Biometrika. 1989;76:297307. 41. Hakkesteegt MM, Wieringa MH, Brocaar MP, Mulder PGH, Feenstra L. The interobserver and test-retest variability of the Dysphonia Severity Index. Folia Phoniatr Logop. 2008;60:8690.