Journal of Voice

Vol. 15, No. 1, pp. 86104

2001 The Voice Foundation

Patterns of Breath Support in

Projection of the Singing Voice
*C. William Thorpe, *Stephen J. Cala, *Janice Chapman, and *Pamela J. Davis
*The National Voice Centre, University of Sydney, Sydney, Australia;
Voice Consultant, London, United Kingdom

Summary: Recordings of the rib cage and abdominal motions and acoustic output
were obtained from five professional opera singers during performance of an aria
recorded with two levels of voice projection. The condition of greater projection resulted in a significant increase in the acoustic power in the frequency band 24
kHz, relative to the power in the 02 kHz band, and a decrease in the mean expiratory flow, implying a move to more efficient vocalization with the greater projection. Also, the condition of greater projection resulted in a larger rib cage, particularly in the lateral dimension, but only a small decrease in the abdominal lateral
dimension, suggesting that the greater abdominal support required for a larger projection is obtained by increased activation of abdominal muscles acting medially.
Key Words: Singing voiceBreath supportChest wall configurationBreathing patternsVoice projectionSingers formantSpectral analysis.

tempts to quantify the physiological characteristics of

support and its effect on the singing voice. The results
of these investigations have however shown a wide
variability in the respiratory strategies employed by
different singers. In Leanderson et al.,5 the use of the
diaphragm in generating support was shown to vary
between singers. Griffin et al.6 showed that although
there was a difference in voice quality between supported and unsupported singing, the breathing patterns were highly variable, and furthermore the
singers perceptions of their own breathing differed
from the empirical measurements. Recently Thomasson and Sundberg7 have shown that professional
singers show consistent breathing patterns when repeating the same music a number of times, implying
that the singers are following some optimized breathing pattern.
In this study we attempt to reduce the intersubject
variability by recruiting singers who have been
trained by a single teacher and so presumably follow
a more consistent method of support than a more

INTRODUCTION
The teaching of singing has developed over many
centuries, but there is still much controversy about
what techniques will optimize singing performance.
Experienced singers and teachers refer to the importance of abdominal or diaphragmatic muscle support
for achieving optimal control of the breath, influencing
tone quality, range, dynamics, and especially projection
of the sound over an orchestra.1,2 However, the nature
of the contribution of abdominal/diaphragmatic muscle
support to vocal control and efficiency (if any) is not
clear, and the methods used to teach breath control for
singing vary widely.3,4 Recently there have been at-

Accepted for publication February 24, 2000.

Address correspondence and reprint requests to Dr. C.W.
Thorpe, School of Communication Sciences and Disorders,
University of Sydney, Post Office Box 170, Lidcombe NSW
1825, Australia.
e-mail: w.thorpe@cchs.usyd.edu.au

86

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

general sample of singers (acknowledging thereby
that our results may be strictly applicable only to this
method). We take the term support to encompass
adjustments in respiratory patterning throughout the
breath cycle, and specifically those made by singers
to facilitate the requisite control of subglottal pressure without compromising the vibratory operation
of the vocal folds. The implication is that the method
used to achieve support is not independent of the
functioning of the vocal folds, so that achieving a
goal such as enhanced projection requires concomitant adjustments to the support mechanism. We
therefore measured the acoustic output and respiratory patterns of singers as they sang with and without
enhanced levels of voice projection. Our hypothesis is that increasing the level of projection should be
accompanied by changes to the respiratory patterns
reflecting the changed use of support.
METHODS
The subjects comprised 5 professional singers, 4 of
whom are or have played leading roles in international-class opera companies (refer to Table 1 for
subject details). One subject was at an early stage in
her professional career (2 years as soloist). Coauthor
JC was herself one subject, and chose the other four
based on her knowledge of their proficiency in her
breath support technique. One subject was therefore
aware of the aims of the experiment, whereas the
other four were given a general explanation that we
were investigating breathing in opera singing. All
procedures were approved by the University and
Hospital Human Ethics Committees.
All subjects had been taught and were proficient in
a specific method of support taught by author JC as
part of a strategy to increase the intensity of projection in the voice quality.8 This method emphasizes

87

the use of abdominal support synchronized with the

onset of phonation. The principal abdominal muscles
actively involved are thought to be transverse abdominis, rectus abdominis, and internal and external
obliques. The support activity can be palpated by the
teacher or student, particularly at three centers of activity: at the level of the xiphoid anchor just below
the sternum; around the lateral abdominal girdle; and
at the pubic junction 13 in. above the pubic bone. In
each of these three areas it is possible to feel the muscles contract by firm palpation at the onset of phonation or fractionally ahead in response to prephonatory
tuning. As part of the process of learning this method,
JC invokes primal sounds (e.g., cry, laugh, yell) that
when produced seem to naturally generate muscle
contractions in the same centers of activity.2,8 JC
teaches that all singing should be supported, but it is
noticeable that during projected singing, there is a
particular increase in muscle contraction in the lateral
abdominal region. This lateral abdominal support appears to provide stability to the actions of the rib cage
and diaphragm during phonation. JC emphasizes that
this support should never be used during inspiration,
but that the abdominal muscle tension should be released at the onset of inspiration to allow a rapid descent of the diaphragm. From the pedagogical experience, it appears that the total release of the abdominal
support during inspiration facilitates the smooth reactivation of support for the subsequent phrase.
Because of the requirement to schedule experiments when subjects were available, it became necessary to make use of two separate recording environments. The first was a respiratory laboratory in a
hospital, where three subjects were recorded, and the
second a television studio where a further three subjects (including a repeat of one subject) were recorded. The reverberation time constant (RT60) was measured as 0.3 s for the laboratory and 0.2 s for the

TABLE 1. Subject Characteristics and Aria Details

Subject

Voice Type

Age/Years Performing

Aria

Opera

Composer

Tenor

43/20

Salut! Demeure chaste et Pure

Faust

Gounod

Soprano

59/35

Vissi darte

Tosca

Puccini

Soprano

26/2

Quando men vo

La Bohme

Puccini

Baritone

35/14

Io morr, ma lieto in core

Don Carlos

Verdi

Tenor

38/9

Nessun dorma

Turandot

Puccini

Journal of Voice, Vol. 15, No. 1, 2001

88

C. WILLIAM THORPE ET AL

television studio. Because of these factors, absolute

sound pressure levels could not be reliably determined across the entire spectrum or compared between subjects. We therefore normalized the sound
levels for each subject to the maximum sound pressure achieved by that singer. Intersubject comparisons can therefore be made based on the limited assumption of relative equivalence between each
subjects maximum sound pressure level. Recording
levels and the microphone to subject distance were,
however, kept constant so that sound levels could be
compared between conditions for each subject.
Sound was recorded on a digital audio tape (DAT)
recorder (DA-P1, Tascam, Montebello, CA) at a
sampling rate of 44.1 kHz, and subsequently transferred to computer via a digital sound interface (Audiomedia II, Digidesign, Palo Alto, CA) and resampled at 22.05 kHz to reduce memory requirements.
Linear magnetometers (GMG Scientific, Burlington, MA) were employed to measure the anteriorposterior (AP) and lateral dimensions of the rib cage
and abdomen. Sensors were attached to the skin with
double-sided tape at the umbilical level for abdominal dimensions and the nipple level (axilla for female
subjects) for the rib cage dimensions. An adjustable
frame was provided for the subjects to rest their arms
on to reduce movement artifacts. The magnetometer
signals were sampled at 100 samples per second
(Maclab, AD Instruments, Castle Hill, NSW Australia) together with a representation of the acoustic
signal for synchronization purposes. The magnetometer signals were calibrated in centimeters so that
the relative movements of the lateral and AP dimensions of the rib cage and abdomen could be compared. For the purposes of lung volume estimation,
calibrations were also performed against spirometric
measurements obtained with a drum spirometer
(Morgan Spiroflow, PK Morgan, Andover, MA)
while performing vital capacity (VC), isovolume,
and relaxation maneuvers. The volume estimate signal was formed from a weighted sum of the two AP
magnetometer signals, with the weighting factor calculated by a least squared error fit to the isovolume
data. The measurements of residual volume (RV) and
total lung capacity (TLC) obtained during the VC
maneuvers were then used to derive a volume scale
of percent VC, with the zero level at RV and 100%
VC representing TLC. This scale was used instead of
absolute volume units to allow for comparisons beJournal of Voice, Vol. 15, No. 1, 2001

tween subjects. Relaxations were obtained through a

high-resistance orifice switched into the expiratory
circuit after the subject had relaxed against a closed
valve at TLC. Three or four coached relaxations were
attempted, and the most consistent two used here.
Subjects sang a part of an aria from their repertoire,
selected in consultation with coauthor JC. The arias
are listed in Table 1. The aria was sung twice, and the
singers were instructed first to perform as if projecting their voices over a large orchestra, and for the
second take, to perform as if with a smaller accompanying orchestra, without such a need to project
their voices. Coauthor JC understood these instructions to mean singing with and without the use of
enhanced projection. For the second group of three
subjects, each condition was repeated so that four
takes in total were performed by these singers. The
order was randomized as shown in Table 2. The
singers were accompanied by a pianist playing an
electronic piano whose output was fed into openfield headphones which allowed the performers to
hear the accompaniment with little disruption to
hearing their own voices.
Individual breaths in the recorded data were numbered for identification and the times of initiation and
termination of vocalization within each breath were
manually located and marked. Figure 1 illustrates the
placement of these markers. It was necessary to refer
to both the respiratory and acoustic data to accurately position these markersas indicated in Figure 1A,
the instants at which vocalization is initiated or terminated within a phrase may not correspond to maxima or minima of the magnetometer signals. On occasion breaths were taken at different locations in the
two takes (i.e., a single breath in one take was replaced by two shorter breaths in the other take), so
these breaths were identified and excluded from the
paired comparisons between the conditions. Table 2
details the total number of breaths in each take and
how many breaths were taken at different locations
between takes. For the subjects from which two takes
of each condition were recorded, the per-phrase parameters were averaged across the two takes before
performing subsequent statistical analyses.
We define here the initiating lung volume (ILV)
and terminating lung volume (TLV) for each individual phrase as the values of the estimated volume signal at the instants where vocalization begins and ends
within that phrase, respectively (refer to Figure 1).

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

89

TABLE 2. Phrasing Details of the Takes, Indicating the Order of Takes Duration, and Number
of Breaths in Each Take*
Number of Breaths
Subject

Order of Takes

NP

Duration of Extracts (s)

Anomalous Breaths

NP

P-P-NP-NP

30, 29

30, 30

2 (phr 21 in takes 1,2,4

phr 27 in takes 2,3)

206, 207

211, 212

NP-P-P-NP

23, 24

24, 24

1 (phr 14 in take 2)

166, 167

163, 166

10

10

69

63

23, 23

23, 23

132, 134

130, 134

2A

P-NP

P-NP-P-NP

P-NP

24

24

133

136

P-NP

22

22

2 (phr 15 in take 1,
phr 11 in take 2)

154

177

Abbreviations: NP, nonprojected; P, projected.

*The anomalous breaths are indicated by the phrases in which two breaths were taken in the other takes in place
of one breath in the indicated take.

Total volume used during vocalization Vexp is defined by the difference ILV TLV, and mean vocalization flow (MVF) is defined as Vexp/Texp, where
Texp is the duration of the phrase vocalization and
Vexp is volume expired. We also measured the inspired volume Vinsp, and duration of inspiration Tinsp,
for each individual phrase. From the magnetometer
signals we obtained the AP and lateral dimensions at
the initiation and termination of vocalization and the
maximum movement extent during each phrase (note
that the maximum extent is not necessarily equal to
the initiation minus termination values since, as indicated in Figure 1, the extremes of movement may occur sometime between the phrase endpoints).
The relative contributions of rib cage and abdomen
to the respiratory cycle are indicated by the KonnoMead diagram,9 which plots the relative movement
of one against the other. Referring to Figure 1B, abdominal movement is indicated on the horizontal axis and rib cage movement on the vertical axis. Outward motion (implying inspiration) is upward and
toward the right. Any breathing pattern can be displayed on the plot by simultaneously tracing the dimensions of the rib cage and abdomen. Reference
points are provided by the relaxation curve, RV, TLC,
and functional residual capacity (FRC) points on that
curve, and the slope of the isovolume line. By scaling the axes such that an isovolume maneuver dis-

plays at a slope of 45 degrees, the axes indicate abdominal and rib cage movements in equal volume
units.10 A straight line was fitted through the raw relaxation rib cage and abdomen displacement data for
presentation on the Konno-Mead diagrams here. Because our subjects were not experienced in the performance of relaxation maneuvers (and the upright
posture imposes some muscle tone anyway), we only made use of the data from the upper part of the
volume range. We present Konno-Mead plots for representative breaths from the data. The path followed
during a breath, in particular its relative position with
respect to the relaxation curve, can implicate different muscle activation patterns.
For each phrase, the mean acoustic power in dB
(unweighted) was computed. We then computed the
average power spectral density by means of the fast
Fourier transform (FFT) calculated on a series of
overlapping segments throughout the duration of the
phrase. The segment size was set to 20 ms (440 samples), with segments spaced at 5 ms. A Blackman
window was applied to each segment which was then
extended to 2048 samples before computing its FFT.
The mean of the squared absolute FFTs was calculated and normalized to take account of the effects of
the window. The power in the average spectrum is
thus comparable to the power obtained from the time
domain acoustic signal. From the average power
Journal of Voice, Vol. 15, No. 1, 2001

90

C. WILLIAM THORPE ET AL
ies that have shown that the acoustic energy or peak
amplitude within the band 24 kHz gives a good representation of the ringing quality in a singers
voice.11-13 The ratio Phi/Plo was also calculated.
RESULTS

B
FIGURE 1. Depiction of the respiratory and acoustic signals,
indicating the points at which initiation and termination of the
phrase are identified. A. Rib cage, abdominal, and estimated volume traces of two breaths, together with the associated acoustic
signal. B. Konno-Mead diagram showing the rib cage dimension
plotted against the abdominal dimension of the breaths shown in
part (A). The phrase initiation and termination points of each
breath are marked as I_ and T_, respectively in both (A) and
(B).

spectral density for each phrase, the power in the frequency bands 02 kHz and 24 kHz was obtained
and denoted by Plo and Phi, respectively. The choice
of these frequency bands was based on previous studJournal of Voice, Vol. 15, No. 1, 2001

Figures 2 and 3 illustrate the respiratory patterns of

the subjects during singing their aria both with and
without projection of their voices. The rib cage and
abdominal AP motion during the loudest phrase is
shown in Figure 2 on Konno-Mead plots, sized so
that the length of each axis equals an equivalent volume contribution of 100% of the subjects VC. Isovolume lines are shown at both RV and TLC lung
volume levels (note that the actual isovolume maneuvers were performed at around FRC, but these
lines indicate the volume extremes of movement defined by the model of breathing based on rib cage and
abdominal movement only. A straight line fit through
the relaxation curve, and the FRC configuration during relaxed quiet breathing, are also indicated for reference purposes. We should point out that time is not
represented on these plots, so that slow and fast
movements that follow the same path are not distinguished. Figure 3 shows the time-course of the magnetometer and lung volume signals for the same
phrases represented in Figure 2, with the lengths of
the vertical axes again set equal to 100% of each subjects VC .
The smoothness or otherwise of the vocalization
paths shown in Figures 2 and 3 is of course affected
by the phrase structure, which differs between subjects because of the differences in their singing repertoire. Despite this, notable similarities between the
patterns shown by different subjects include the inspiration path, which in nearly all cases begins with
a rapid isovolume movement in which a rapid expansion of the abdomen occurs concurrently with a
decrease in rib cage dimension such that there is negligible net change in volume. It should be noted that
the duration of this initial part of the inspiration path
is of the order of 100 ms, as indicated by the tracings
shown in Figure 3. The inspiration then continues,
with both rib cage and abdominal movement following the path of the relaxation line, after which a further realignment occurs involving a simultaneous elevation of the rib cage and drawing in of the

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

91

FIGURE 2. Konno-Mead plots for the loudest phrase from each subject. The two
traces represent the phrase in the projected (dark line) and nonprojected (gray line)
conditions; the inspiratory portion is shown dotted. The text, taken from the score,
identifies the phrase. Each plot is scaled so that the isovolume lines at RV and TLV
occur in the same position relative to the relaxation line, with the tick marks on the
axes representing 25%VC volume increments. For clarity, labels are only applied to
the left lower chart, but all others are exactly equivalent. Note that the graphs labeled
Subject 2 and 2A refer to the same singer, but 2A was recorded 1 year earlier.

abdomen. Vocalization is often initiated part-way

through this change, after which it continues with a
simultaneous decrease in both rib cage and abdomen

dimensions, so that the respiratory path followed is

roughly parallel to the relaxation line. There do not
appear to be consistent changes in the paths followed
Journal of Voice, Vol. 15, No. 1, 2001

92

C. WILLIAM THORPE ET AL

FIGURE 3. Time waveforms of the rib cage and abdomen AP movements, and resulting
lung volume estimate, for the phrases shown in the Konno-Mead plots of Figure 2 for each
subject. The dark line represents the projected condition and the gray line the nonprojected condition. The horizontal line on each graph indicates the FRC configuration, while
the instants at which vocalization begins are indicated by a vertical line. Note that for Subject 1 there is an additional short inspiration for the projected condition only. As in Figure
2, only the lower left chart is annotated.

between the projected and nonprojected conditions,

although this could simply be because of the musical
differences between phrases shown for each subject.
Journal of Voice, Vol. 15, No. 1, 2001

It is worth noting that the two plots for Subject 2 (2

and 2A) in Figure 2, representing the same phrase
sung by the same singer, were obtained a year apart.

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

Although the overall shape is similar, there is quite a
difference in the detailed shape at the pitch transition
in the phrase, with a sharp movement upward and to
the left on one occasion (2A) but a smoother (although convoluted) movement to the left on the other occasion (2). When the temporal tracings of the
signals (Figure 3) are examined, this difference is
seen to occur because the rib cage movement is later,
and slightly larger, on the earlier occasion (2A).
Figure 4 shows average power spectra obtained
from each singer for the two conditions. Male singers
are on the left and female on the right, highlighting the
prominent singers formant in the male voice. However, it is evident that the energy in the higher frequency
regions is greater for the projected condition in both

93

male and female singers (although it is clear that there

is not just a single formant that is enhanced). The
differences between Subject 2 and 2A (the same
singer) are partly because only a portion of the aria
was recorded on occasion 2A (including the most dramatic phrases, which possibly accounts for the greater
observed difference). Only two singers (Subjects 4 and
5) show some increased energy within the 0-2 kHz
band for the projected case. The difference between
the projected and non-projected conditions, averaged
across all subjects, is shown in Figure 5, indicating that
there is increased acoustic energy throughout the frequency range 1.66 kHz (although the average energy
in the singing voice falls off rapidly above 34 kHz as
shown by the curves in Figure 4).

FIGURE 4. Average power spectral density over all breaths for each condition shown
for each of the subjects. The solid line represents the projected condition and the dotted line the nonprojected condition. Each graph has the same axis dimensions and labels as identified on the lower left chart.
Journal of Voice, Vol. 15, No. 1, 2001

94

C. WILLIAM THORPE ET AL

FIGURE 5. Average phrase-by-phrase difference in power spectral density between the projected and nonprojected conditions.

Descriptive parameters extracted from the respiratory and acoustic data are presented in Figures 6 and
7, respectively, with a summary of the overall statistics appearing in Table 3. Because much of the wide
spread in parameters indicated in the overall statistics
is a consequence of differences in the musical demands of individual phrases, it is appropriate to perform a phrase-by-phrase comparison between the
projected and nonprojected conditions. The scattercharts shown in Figures 6 and 7 represent the parameter for each phrase as a point having a position
specified by the parameters value in both the projected (horizontal axis) and nonprojected (vertical
axis) conditions. The line of identity, corresponding
to the null hypothesis when both conditions yield
identical parameter values, is also drawn. A preponderance of points on one side of the identity line reveals a shift in that parameter between the two conditions, even when the range of values overlaps
considerably as shown.
Statistical analyses of the phrase-by-phrase comparisons (Table 3) reveal that ILV is not significantly
different between the two conditions, but TLV is significantly higher for the projected condition, and
hence Vexp and MVF are significantly lower. The
phrase durations Texp do not differ significantly, but
Journal of Voice, Vol. 15, No. 1, 2001

Tinsp is significantly faster in the projected condition.

In the acoustic parameters, there is a significant increase in the acoustic power when the singers employ greater projection, but this increase is almost entirely in the high-frequency band, with the average
ratio Phf/Plf increasing by 2.4 dB.
As illustrated in Figure 8, there are some differences between subjects, both in their average parameter values, and in the amount of change consequent on the change in condition. However, the
overall trends across conditions in the respiratory parameters were opposed by only one or two subjects,
in particular Subject 5 who had shorter phrases and
hence higher MVF with little change in Vexp, and
Subject 3 who slightly decreased TLV and increased
Vexp, Vinsp, and MVF. Subject 2 showed consistency
between the two experiment occasions except for
ILV and TLV, both of which decreased with projection on the first occasion (2A) only. Subject 4 increased Plf and Phf almost equally with projection, resulting in a minimal change to Phf/Plf. Notably, all
subjects decreased Tinsp and increased both Phf and
Phf/Plf with projection.
Although there is a change in both the acoustic output and some of the respiratory parameters between
the two conditions, there is little correlation between

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

95

TABLE 3. Statistics of Measured Parameters Over All Breaths for

All Subjects (Excluding Repeat of Subject 2)*
Overall Value
Parameter

Mean

SD
12.1
22.3
20.1
20.5

Projected Nonprojected Difference

Mean
1.9
5.6
3.8
4.3

SD

P(t < T)

7.2
10.3
9.7
12.8

115
115
115
107

2.8
5.8
4.3
3.5

<0.01
<0.0001**
<0.0001**
<0.001**

ILV
TLV
Vexp
Vinsp

(%VC)
(%VC)
(%VC)
(%VC)

80.4
36.0
44.4
43.2

MVF
Texp
Tinsp

(%VC/s)
(s)
(s)

8.7
5.4
0.90

3.4
2.6
0.77

0.71
0.07
0.08

2.8
0.52
0.19

114
115
107

2.7
1.5
4.4

<0.01
NS
<0.0001**

PTOT
PLF
PHF
PHF/PLF

(dB)
(dB)
(dB)
(dB)

26.3
28.2
32.9
4.69

6.0
5.6
8.3
4.8

1.1
0.57
3.0
2.4

2.2
2.0
2.5
2.4

115
115
115
115

5.4
3.1
12.5
11.0

<0.0001**
<0.01
<0.0001**
<0.0001**

Abbreviations: NS, not significant; SD, standard deviation.

* The

significance of differences between the conditions is computed with the paired t-test. Note that because
multiple comparisons are performed, only differences with an individual significance P < 0.002 are significant
at an experimentwide level of P = 0.05. These differences are indicated by **.

relevant acoustic and respiratory parameters, as

shown by the scatter plots in Figure 9. There is some
suggestion of a negative correlation between MVF
and Phf/Plf (r 2 = 0.2) for both conditions, implying
that the higher values of this acoustic ratio are associated with more efficient glottal vibration patterns
(e.g., shorter open quotient and closing transient).
To quantitate the contributions of the rib cage and
abdominal compartments, we gave parameters to
each breathing cycle by measuring, with reference to
the relaxed FRC state, the positions on the KonnoMead diagram of the start and end of vocalization, and the minimum and maximum extents of the
curve in each dimension. Equivalent measurements
were also performed on the lateral dimensions. These
parameters were then subjected to paired t-tests between the projected and nonprojected conditions. Results of these comparisons appear in Table 4. Consistent with the lung volume results shown in Table 3,
there is little change in the AP dimensions. Notably,
there is a significant increase in the RC-AP dimension, particularly at phrase termination, indicating
that the previously observed increase in TLV is obtained largely by an increase in rib cage dimension.
Journal of Voice, Vol. 14, No. TBA, 2001

However, the lateral dimension of the rib cage is significantly wider for the projected condition at both
initiation and termination of the voice, and the lateral abdominal dimension is significantly narrower at
initiation. Note that although the lateral dimension of
the rib cage was only measured for three subjects,
the significance levels of the other measurements are
only slightly changed if they are also computed for
these three subjects only.
DISCUSSION
Although the use of correct breath support is generally regarded as necessary in order to effectively
produce the singing voice, the relationship between
breath support and voice projection is still little understood. Our findings in the experiments described
here provide some insight into this relationship.
Measurement accuracy
It is relevant to first comment on the measurement
accuracies of volume estimates based on rib cage and
abdominal dimensions. These are predicated on a linear relationship between the measured dimension
Journal of Voice, Vol. 15, No. 1, 2001

96

C. WILLIAM THORPE ET AL

FIGURE 6. Scatter charts illustrating the breath-by-breath comparison

of the respiratory parameters between the projected and nonprojected
conditions. The parameters are: ILV = Initiation Lung Volume; MVF =
Mean Vocalization Flow; Texp = expiratory duration; Tinsp = inspiratory
duration; TLV = Termination Lung Volume; Vexp = Volume expired. Each
point on the charts represents the parameter value obtained from a particular phrase in the aria, its (x, y) coordinates on the chart specified by
the pair of values obtained in the projected and nonprojected conditions
respectively. Individual singers are represented by the following symbols:
, Subject 1; , Subject 2; , Subject 2A; , Subject 3; , Subject 4;
, Subject 5. The line of identity, where both the projected and nonprojected values are equal, is also shown on each chart.

Journal of Voice, Vol. 15, No. 1, 2001

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

97

FIGURE 7. Scatter charts illustrating the breath-by-breath comparison of

the acoustic parameters PLF = power in 02 kHz band, PHF = power in 2-4
kHz band, PTOT = total acoustic power, and PHF/PLF = ratio of power between
high-frequency and low-frequency bands. Individual singers are represented
by the following symbols: , Subject 1; , Subject 2; , Subject 2A; , Subject 3; , Subject 4; , Subject 5.

and the internal volume, with an assumption that two

compartments are sufficient.9 It is known however
that this model loses accuracy at high thoracic pressures or with movement of the torso.14,15 Although
we attempted to reduce extraneous movements by
supporting the singers arms, the high pressure
swings inherent in singing activities mean that the
volume estimates are limited in accuracy. Furthermore, our measurements of the lateral dimensions
showed that significant movements are not always
correlated with the AP dimensions, indicating that
the AP measurements alone may not be sufficient to
represent all changes in torso volume. Particularly at
points in the breath cycle when there are large
changes in thoracic pressure, it seems likely that the

chest wall shape is distorted to an extent that the simple relationship between AP dimension and volume
is lost. Such events can be observed in several of the
results shown in Figures 2 and 3 where there are momentary increases or decreases in estimated lung volume associated with sudden motions of the rib cage
and abdomen (in most cases occurring at points
where there are large changes in musical dynamics).
Possible inaccuracies in the lung volume estimation are also implied by the results that apparently
show certain singing breaths extending beyond the
TLC and RV limits (e.g., refer to Figures 2 and 6).
This may be due to the effect that breathing on a
mouthpiece (as during the calibration) has on breathing behavior,16 but also because the TLC and RV
Journal of Voice, Vol. 15, No. 1, 2001

98

C. WILLIAM THORPE ET AL
conditions, the absolute accuracy limitations of the
volume measurements do not overly detract from the
results. Rather, it is the differences between the conditions that are important in our interpretations. We
assume that any measurement distortions are similar
between the two conditions.

FIGURE 8. Individual subject averages for each parameter, shown in both nonprojected (NP) and projected
(P) conditions. The parameters are as in Figures 6 and
7. Individual singers are represented by the following
symbols: , Subject 1; , Subject 2; , Subject 2A; ,
Subject 3; , Subject 4; , Subject 5.

limits were obtained during somewhat artificial respiratory maneuvers that may not reflect the true capabilities of the singers when naturally engaging
their trained respiratory responses during singing.
However, because our experiment is designed so that
we perform paired comparisons between the two
Journal of Voice, Vol. 15, No. 1, 2001

Lung volumes
It is perhaps surprising that breathing patterns
changed so little when the singers increased the
intensity of their voice projection. Previous studies,13,17-19 have indicated a relationship between the
sound intensity of speech and singing and the lung
volumes used, in particular that at higher intensities
both speakers and singers tend to breath at higher
lung volumes. In contrast, our results show that
breaths are not initiated at higher lung volumes when
the voice intensity is increased in the projected condition. The ILV values during singing an aria are of
course much higher than in speech and quieter
singing,20 so one would not expect that much of an
increase was possible. There is, however, a significant decrease in the average flow rate, and hence
breaths terminate at higher lung volumes.
Our finding that the initiation volumes of the
phrases did not change even when the acoustic output increased perhaps indicates the extent to which
the breathing of these singers is optimized for the requirements of the particular music. The sizes of the
breaths for each phrase are set by the requirements of
the music, and must be learned sufficiently well so
that the singer can automatically take in a breath of
the required size for each phrase during a performance. Watson and Hixon (1996)20 showed that
there was a strong training effect when one singer became more familiar with a particular aria. Because
our subjects all sang arias that were part of their performance repertoire, we can assume that they always
took the size of breath required for each phrase, as
determined by their previous practice and performance. This consistency between different takes was
also observed in the study of Thomasson and Sundberg,7 and its appearance in our data, even though we
instructed our subjects to sing with different levels of
projection, confirms that the memory of the inspiration required, which evolves during the training period, is strongly fixed in well-learned arias. Indeed,

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

99

FIGURE 9. Correlations between respiratory parameters (ILV = Initiation Lung Volume and MVF = Mean Vocalization Flow) and acoustic parameters (PTOT = total power and PHF/PLF = ratio of power between high-frequency and low-frequency bands.

singers often say that they only sing a piece in one

dynamical manner according to its musical and emotional requirements. The consistency of breathing
patterns affirms the singers perception, but the observed increase in acoustic output in the singers formant suggests that subtle alterations in the singing
output are indeed possible in response to appropriate
stimuli.
Acoustic output
It is pertinent to note that the increase in acoustic
output we observed is predominantly comprised of
an enhancement in the frequency band 24 kHz, with
only a marginal increase in sound energy in the frequency band 02 kHz. Indeed, some negative correlation occurs between the average flow rate for each
phrase and the high-frequency/low-frequency acoustic power ratio. This implies that there is a move to-

ward more efficient vocalization in the projected

condition. Griffin et al.6 also found that the glottal
open quotient was lower when their singers were using a supported voice. As described by Sundberg,21
there is a tendency for higher frequencies to be enhanced as sound intensity increases, simply because
of the nonlinearity of the glottal flow waveform.
However, the actual decrease in MVF suggests an alteration in the laryngeal configuration when the
singers are asked for greater projection, with the
shift in acoustic energy to the high-frequency band
caused by a steepening of the glottal closure transient
as the glottal open quotient reduces.22,23
Several other studies have utilized the long-term
average spectrum (LTAS) to quantify the level of the
singers formant. Rossing et al.11 compared the
acoustic characteristics of experienced soprano
soloists who sang similar passages as if in a choir
Journal of Voice, Vol. 15, No. 1, 2001

100

C. WILLIAM THORPE ET AL
TABLE 4. Statistics of the Four Dimensions Across All Breaths from All Subjects*
Overall Value
Variable

Mean

SD

Projected Non-projected Difference

Mean

SD

P(t < T)

RCAP-I

(cm)

1.05

0.59

0.07

0.23

115

3.3

0.001**

RCAP-T

(cm)

0.34

0.54

0.11

0.26

115

4.5

<0.0001**

RCAP-EXT

(cm)

1.08

0.75

0.01

0.20

115

0.4

RCLAT-I

(cm)

0.17

0.19

0.19

0.27

76

6.0

<0.0001**

RCLAT-T

(cm)

0.12

0.18

0.19

0.27

76

6.1

<0.0001**

RCLAT-EXT

(cm)

0.46

0.36

0.01

0.19

76

0.44

NS

ABAP-I

(cm)

0.64

0.48

0.07

0.36

115

2.1

<.05

ABAP-T

(cm)

0.72

0.68

0.04

0.28

115

1.7

NS

ABAP-EXT

(cm)

1.45

0.59

0.08

0.35

115

2.4

ABLAT-I

(cm)

0.06

0.28

0.11

0.19

115

6.1

<0.0001**

ABLAT-T

(cm)

0.05

0.22

0.01

0.13

115

1.2

NS

ABLAT-EXT

(cm)

0.38

0.28

0.06

0.20

115

3.3

NS

<0.05

0.001**

*Note that RCLAT was not obtained for two subjects. Dimensions of the rib cage (RC) and abdomen (AB) in
the anterior-posterior (AP) and lateral (LAT) axes are shown relative to the FRC configuration at initiation
(-I) and termination (-T) of vocalization, together with the maximum movement extent (-EXT) along each axis.

and as a soloist. They found that with solo

singing there was a significant increase in the energy of spectral peaks around 3 kHz in the LTAS suggestive of an increased singers formant. They also
measured increased energy in the frequency band
24 kHz. Omori et al.12 characterized the ringing
quality of a singers voice by the ratio between the
greatest harmonic peak in the 24 kHz band to that in
the 02 kHz band. They found a significantly greater
ratio for professional singers than for nonprofessionals or nonsingers. Ekholm et al.13 examined the relative energy in the region of the singers formant
peak identified in the LTAS (choosing a bandwidth
defined by the trough between this peak and lower
harmonics, which for our male singers at least is approximately at 2 kHz). They compared this measure
to perceptual ratings of the voice, finding a high correlation with ratings of color/warmth and resonance/ring. In combination with these studies, our
results also suggest that the relative energy in the
Journal of Voice, Vol. 15, No. 1, 2001

band 24 kHz in the LTAS provides a good measure

of what singers achieve by increasing the projection
characteristics of their voices, which may be related
to qualitative terms such as ring or twang.
It seems likely that the relationship between vocal
loudness and lung volume may be related to the way
in which the increased loudness is elicited. In a study
by Stathopoulos and Sapienza (1993)17 in which a
positive relationship between speech intensity and
lung volume was observed, speech intensity was controlled by means of visual feedback of sound pressure level (SPL). By contrast, Winkworth and Davis
(1997)24 employed the Lombard effect (i.e., an increase in ambient sound level) to increase speech intensity without explicitly mentioning SPL to the subjects. They found no relationship between speech
intensity and lung volume. In our study likewise, the
instructions to the subjects were to imagine that they
were projecting their voices over a large or small
orchestra. It is possible that this indirect elicitation of

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

increased vocal output is associated with a different
physiological process than is invoked by a direct conscious demand for more sound.
Chest wall configuration
during the singing breath cycle
The Konno-Mead plots of the rib cage versus abdominal motions are interesting with respect to their
correspondence to the pedagogical instructions. Indeed, it is remarkable that all five singers showed a
similar overall pattern. When we examine the inspiratory part of the breath, there is an initial rapid outward movement of the abdomen, often associated
with a momentary inward motion of the rib cage.
This paradoxical pattern was also noticed by Watson
and Hixon20 in their study of classical singers. The
short duration of the paradoxical motion, on the order of 100 ms as indicated in Figures 1A and 3, implies that it is simply a mechanical effect of the decompression of the thoracic cavity as the expiratory
muscles are released and the diaphragm activated for
inspiration. This action appears to correspond to the
splat maneuver taught by Chapman, where she encourages a rapid deactivation of abdominal support
musculature together with the diaphragmatic contraction (resulting in a rapid outward movement of
the abdomen). The word splat is sometimes used to
remind the singer to rapidly release the abdominal
tension. The release of the support muscle activity, in
combination with the negative thoracic pressure generated by the rapid diaphragm contraction, results in
the rib cage quickly returning to its relaxation position. This is what produces the rapid paradoxical
isovolume motion observed at the beginning of most
of the fast inspirations in the data.
It seems reasonable to suggest that the splat maneuver, which combines a cessation of all support
muscle activity (presumably including both inspiratory and expiratory muscle groups) with a rapid and
strong activation of the diaphragm, is necessary to
achieve the rapidity and depth of inspiration that is
necessary for this type of singing. Even in the data of
Watson and Hixon20 where a variety of patterns are
exhibited, comparable patterns to what we observed
are seen in some of their more experienced singers,
notably in the aria and fast singing. The opposite
pattern, of inspiration via an increase in rib cage dimension with the abdomen held in, was most often

101

observed by them in slow singing in one subject who

apparently came from a different pedagogical tradition than the other subjects. In our data, the smallest
extent of the splat pattern was seen in the singer
(Subject 3) with the least experience.
After the splat event, the inspiration continues
with both the rib cage and abdominal compartments
expanding, roughly along the relaxation line. This
implies that the rib cage and abdomen compartments
fill in proportion to their relative relaxed compliances, which implies an efficient allocation of inspiratory muscle action to the two compartments. With
respect to the pedagogy, this part of the inspiration
corresponds to the all round expansion of the rib
cage and belly described by Chapman and other
practitioners. As shown in the detailed results, there
is a wide spread in the inspiratory durations, ranging from 0.2 to 4.4 s (note that musical pauses were
not counted as part of an inspiration). There was a
significant decrease in the inspiratory duration in
the projected condition with all subjects, which
could be a side effect of the slightly smaller volume
requirements, or perhaps because this condition was
closer to the performance situation that the singers
were used to, thus facilitating a more natural, faster,
inspiration.
At the end of inspiration, the chest wall again
moves away from the relaxation line, with an inward
movement of the abdomen coupled with a continued
outward motion of the rib cage. However, the increase in lung volume at this point is negligible, and
again the movement can be seen as a paradoxical isovolume maneuver to prepare for the subsequent vocalization. Other studies have examined this
prephonatory posturing in vocalization,10,25,26
with the consensus being that this movement away
from the relaxation configuration optimizes the chest
wall mechanics for phonation. This phase corresponds to the activation of support, obtained by a
drawing in of the abdomen to create the pressure required for vocalization. Indeed, the results shown in
Table 4 indicate that the rib cage is raised from its relaxed position throughout the vocalization, and to a
greater extent in the projected condition. The abdominal AP dimension is somewhat greater than the relaxed position at the beginning of vocalization, but
moves to a smaller size at the end. There is no difference with the two projection conditions.
Journal of Voice, Vol. 15, No. 1, 2001

102

C. WILLIAM THORPE ET AL

Significantly, the lateral dimension of the rib cage

increased, and that of the abdomen decreased, in the
condition of increased projection. Both changes were
greater than any shift in AP dimension, implicating a
distortion of the chest wall shape brought about by a
change in the activation pattern of support. The dimension changes are suggestive of an increased abdominal pressure with the greater support invoked
for the projected condition. Because the lower part of
the rib cage apposes the abdominal compartment, an
increased abdominal pressure exerts an increased
outward force on this part of the rib cage,27 which
could account for the increase in the lateral dimension of the rib cage during the projected condition.
Likewise, strong activation of the transverse and
oblique abdominal muscles would tend to compress
the circumference of the abdomen and therefore narrow its lateral dimension, as occurs in our results.
This finding concurs with the practice of teachers
such as author JC who train singers in the actions required for support by asking them to feel with their
hands the increased tension in the lateral regions of
their abdomen when they successfully engage the
support.
The phonatory posturing maneuver can also occur during vocalization when there is a need for a
sudden increase in vocal output (either SPL or fundamental frequency) during a phrase, as indicated in
Figure 2 by the patterns for several of the subjects.
For instance, Subjects 1 and 2 both require a significant increase in subglottal pressure part way through
the phrases illustrated in order to attain large pitch
leaps. The opposite pattern, when sudden brief cessation of vocalization requires a rapid release and reactivation of support, is shown by the phrase of Subject 3 in Figure 2. In general it appeared that the
inward movement was accompanied by an elevation
of the rib cage as the pleural and abdominal pressures
increase. However, if expiratory muscles in the rib
cage are activated simultaneously or earlier than the
abdominal contraction, the resulting movement can
in fact be reduced as indicated in the two recordings
obtained (1 year apart) for Subject 2, shown in Figures 2 and 3. It appears that on the later occasion that
muscles driving the rib cage movement may have
been activated somewhat earlier (and hence more in
synchrony with the abdominal muscle activation) so
that the outward movement in response to the conJournal of Voice, Vol. 15, No. 1, 2001

traction of the abdomen was significantly reduced. If

pressures had been measured, one would have expected that the increase in subglottal pressure would
have been delayed on the earlier recording because of
the transient movement of volume between the abdomen and rib cage, but more rapid on the second occasion due to the simultaneous activation of rib cage
and abdominal muscles. The implication of this observation for singers is that simultaneous activation
of rib cage and abdominal muscles may result in
more rapid and possibly better controlled changes in
subglottal pressure. Note that for Subject 1 the rib
cage movement occurs even earlier (with respect to
the abdominal motion) than in Subject 2, although
this difference may be partly due to the /s/ between
the two notes. Wilder26 suggests that different patterns of prephonatory motion may occur because of
differences in posture or in the biomechanical properties of the chest wall between subjects. Both effects
could also have contributed to the observed change
for Subject 2 between the two recordings, in addition
to possible changes in technique over the intervening
year, as there are differences in the relaxation curve
and FRC configuration which imply some biomechanical change. Overall, however, the resulting parameter values for this subject are very similar for the
two occasions (Figure 6).
Vocalization often begins during the movement
away from the relaxation line, particularly when the
vocal output required is relatively low for the start of
the phrase. However, subsequent to the activation of
support, sustained vocalization occurs with both the
rib cage and abdomen moving inward at a remarkably similar rate (relative to the isovolume and relaxation references). The consistency between subjects
seems to be greater than in some previous studies
where a range of breathing patterns was found,7,20
probably due to our selection of subjects who follow
the technique of a single teacher. Only one singer
(Subject 3) shows much greater rib cage than abdominal contribution to the volume expired, but in all
subjects the motion is roughly parallel to the relaxation line, suggesting that (during a sustained vocalization) the rib cage and abdominal muscles are activated jointly so that the respiratory system acts as a
single compartment. This could provide a gain in efficiency or perhaps improved control of the sustained
high pressure required for such vocalizations. This

BREATH SUPPORT IN PROJECTION OF THE SINGING VOICE

pattern is in contrast to some of those observed in
other studies20,28 where patterns of either predominantly rib cage or abdominal motion were also observed. Indeed, some pedagogical traditions teach
different patterns such as holding the rib cage out
throughout a vocalization, with abdominal movement providing the volume change.1
CONCLUSIONS
In conclusion, our findings tend to reinforce the
observation of many teachers and singers that good
support provides needed assistance in projecting the
voice. The use of support exhibits itself here as a
movement away from the relaxation state, with abdominal muscle activation and a raising of the rib
cage, coupled with a rapid release of expiratory muscle activity at the start of inspiration. With the request
for an increased level of projection, the rib cage was
raised slightly further, and there was a significant
shift in the lateral dimensions of the abdomen and rib
cage that was suggestive of an increased activation of
the abdominal muscles in the lateral area. However,
the lung volume at phrase initiation was remarkably
consistent, suggesting a well-developed muscle
memory of the inspiration required for each phrase.
With respect to the increased acoustic output when
the subjects were asked to project as if over a large
orchestra, it is interesting that the additional energy
is indeed concentrated in the region of the singers
formant. This, together with the associated reduction
in flow, suggests that the increased projection is accomplished by an alteration in the glottal vibratory
pattern. Based on other experimental5,22,29 and analytical30 studies, it seems likely that the condition of
greater support leads to an increase in subglottal
pressure, which, together with simultaneous adjustments to the laryngeal musculature, gives rise to the
changed glottal flow waveform. The mechanisms by
which the integration between the respiratory and laryngeal control is effected still needs some elucidation, however.
Acknowledgments. This work was supported by grants
from the University of Sydney and the Australian Research Council. The authors thank the subjects for their
willingness to partake in the experiments, and Grace Edwards for providing accompaniment.

103

An initial version of this paper was presented at the 27th

Symposium of the Voice Foundation, Philadelphia, PA,
June 1998.

REFERENCES
1. Vennard W. Singing: The Mechanism and the Technic. New
York: Carl Fischer, 1967.
2. Brown OL. Discover Your Voice. San Diego, Calif: Singular
Publishing Group, 1996;
3. Miller R. On the Art of Singing. New York: Oxford University Press, 1996.
4. Spillane KW. Breath Support directives used by singing
teachers: a Delphi study. The NATS. 1989(January/February):9-57.
5. Leanderson R, Sundberg J, von Euler C. Role of diaphragmatic activity during singing: a study of transdiaphragmatic
pressures. J Appl Physiol. 1987;62(1):259-70.
6. Griffin B, Woo P, Colton R, Casper J, Brewer D. Physiological characteristics of the supported singing voice. A preliminary study. J Voice. 1995;9(1):45-56.
7. Thomasson M, Sundberg J. Consistency of phonatory
breathing patterns in professional operatic singers. J Voice.
1999;13(4):529-541.
8. Chapman J. Making connectionsprimal singing. In Proceedings of Voice Conservation Symposium. London: The
Voice Research Society, 1986.
9. Konno K, Mead J. Measurement of the separate volume
changes of rib cage and abdomen during breathing. J Appl
Physiol. 1967;22(3):407-422.
10. Hixon TJ, Goldman MD, Mead J. Kinematics of the chest
wall during speech production: volume displacements of the
rib cage, abdomen, and lung. J Speech Hear Res. 1973;16:
78-115.
11. Rossing TD, Sundberg J, Ternstrom S. Acoustic comparison
of soprano solo and choir singing. J Acoust Soc Am. 1987;
82(3):830-836.
12. Omori K, Kacker A, Carroll LM, Riley WD, Blaugrund
SM. Singing power ratio: quantitative evaluation of singing
voice quality. J Voice. 1996;10(3):228-235.
13. Ekholm E, Papagiannis GC, Cagnon FP. Relating objective
measurement to expert evaluation of voice quality in Western classical singing: critical perceptual parameters. J Voice.
1998;12(2):182-196.
14. McCool FD, Loring SH, Mead J. Rib cage distortion during
voluntary and involuntary breathing acts. J Appl Physiol.
1985;58(5):1703-1712.
15. McCool FD. Noninvasive methods for measuring ventilation. In Roussos C, ed. The Thorax, Part B: Applied Physiology, 2nd ed. New York: Marcel Dekker, 1995:1050-1069.
16. Paek D, McCool FD. Breathing patterns during varied activities. J Appl Physiol. 1992;73(3):887-893.
17. Stathopoulos ET, Sapienza C. Respiratory and laryngeal
function of women and men during vocal intensity variation. J Speech Hear Res. 1993;36(1):64-75.
Journal of Voice, Vol. 15, No. 1, 2001

104

C. WILLIAM THORPE ET AL

18. Dromey C, Ramig LO. The effect of lung volume level on

selected phonatory and articulatory variables. J Speech
Lang Hear Res. 1998;41(3):491-502.
19. Winkworth AL, Davis PJ, Ellis E, Adams RD. Variability
and consistency in speech breathing during reading: lung
volumes, speech intensity, and linguistic factors. J Speech
Hear Res. 1994;37:535-556.
20. Watson PJ, Hixon TJ. Respiratory kinematics in classical
(opera) singers. J Speech Hear Res. 1985;28:104-122.
21. Sundberg J. The source spectrum in professional singing.
Folia Phoniatr. 1973;25(1):71-90.
22. Gauffin J, Sundberg J. Spectral correlates of glottal voice
source waveform characteristics. J Speech Hear Res. 1989;32:
556-565.
23. Sundberg J, Titze I, Scherer R. Phonatory control in male
singing: a study of the effects of subglottal pressure, fundamental frequency, and mode of phonation on the voice
source. J Voice. 1993;7(1):15-29.

Journal of Voice, Vol. 15, No. 1, 2001

24. Winkworth AL, Davis PJ. Speech breathing and the Lombard effect. J Speech Hear Res. 1997;40(1):159169.
25. Baken RJ, Cavallo SA, Weissman KL. Chest wall movements prior to phonation. J Speech Hear Res. 1979;22(4):
862-871.
26. Wilder CN. Chest wall preparation for phonation in female
speakers. In Bless DM, Abbs JH, eds. Vocal Fold Physiology: Contemporary Research and Clinical Issues, San Diego,
Calif: College Hill Press, 1983:109-123.
27. Chihara K, Kenyon CM, Macklem PT. Human rib cage distortability. J Appl Physiol. 1996;81(1):437-447.
28. Watson PJ, Hixon TJ, Stathopoulos ET, Sullivan DR. Respiratory kinematics in female classical singers. J Voice.
1990;4(2):120-128.
29. Schutte, H. K. The Efficiency of Voice Production. Groningen University Hospital, 1980.
30. Titze IR, Sundberg J. Vocal intensity in speakers and
singers. J Acoust Soc Am. 1992;91(5):2936-2346.