Substyles of Belting: Phonatory and Resonatory

Characteristics

n, and ‡Lisa Popeil, *yStockholm, Sweden and zLos Angeles, California
*,†Johan Sundberg, †Margareta Thale

Summary: Belting has been described as speechlike, yell-like, or shouting voice production commonly used in
contemporary commercial music genres and substantially differing from the esthetic of the Western classical voice
tradition. This investigation attempts to describe phonation and resonance characteristics of different substyles of belt-
ing (heavy, brassy, ringy, nasal, and speechlike) and the classical style. A professional singer and voice teacher, skilled
in these genres, served as the single subject. The recorded material was found representative according to a classification
test performed by an expert panel. Subglottal pressure was measured as the oral pressure during the occlusion for the
consonant /p/. The voice source and formant frequencies were analyzed by inverse filtering the audio signal. The
subglottal pressure and measured flow glottogram parameters differed clearly between the styles heavy and classical
assuming opposite extremes in most parameters. The formant frequencies, by contrast, showed fewer less systematic
differences between the substyles but were clearly separated from the classical style with regard to the first formant.
Thus, the differences between the belting substyles mainly concerned the voice source.
Key Words: Formant frequencies–Flow glottogram–NAQ–Subglottal pressure.

INTRODUCTION Moreover, she found that the contact quotient,a as measured

The term ‘‘belting’’ voice refers to the speechlike, yell-like, or
shouting voice production heard in such commercial vocal sub-
styles as pop, rock, R&B, jazz, country, and world music as well
as in musical theater. It differs substantially from the esthetic of
the Western classical voice tradition.
Traditionally, the belting voice has been described as loud,
nasal, and more akin to shouting than to the operatic esthetic
of singing. Over time, this sound, as it is used in the professional
musical theater world, has evolved into a number of discernible
substyles, which might be termed ringy, brassy, nasal, speech-
like, and heavy belts.1 In musical theater singing, each substyle
represents and enhances the portrayal of characters on the stage.
The substyles are commonly used also in other types of so-
called contemporary commercial music.
The need for a scientific description of belting was pointed
out by Hollien and Miles.2 Physiological findings differentiat-
ing classical and belting voice production have shown many
important differences including variances in vocal fold vibra-
tional patterns, speed quotients, and closed quotients of the
vocal folds, air pressure, cartilage tilting differences of the
thyroid and cricoid cartilages, the position of the hyoid bone,
involvement of the supralaryngeal muscles, resonance charac-
teristics, and comparative laryngeal height.
In a single subject investigation, Estill3 analyzed acoustical
and physiological characteristics of belting as compared with
‘‘speech’’ and ‘‘opera’’ modes. She found that belting was
considerably louder than these styles and also that it was pro-
duced with a high degree of EMG activity in the vocalis muscle.
Accepted for publication October 6, 2010.
This investigation was first presented at the Annual Symposium, Care of the Professional
Voice, Philadelphia, June 2009.
From the *Department of Speech Music Hearing, School of Computer Science and
Communication, KTH, Stockholm, Sweden; yUniversity College of Music Education in
Stockholm (SMI), Stockholm, Sweden; and the zLos Angeles, California.
Address correspondence and reprint requests to Johan Sundberg, Department of Speech
Music Hearing, School of Computer Sciences and Communication, KTH, Stockholm
SE-10044, Sweden. E-mail: pjohan@speech.kth.se
Journal of Voice, Vol. 26, No. 1, pp. 44-50
0892-1997/$36.00
Ó 2012 The Voice Foundation
doi:10.1016/j.jvoice.2010.10.007
from an electroglottogram (EGG), was highest in belt and low-
est in opera. She also noted differences in the electromyogram
(EMG) signals recorded from a number of extrinsic laryngeal
muscles.
Schutte and Miller4 compared ‘‘nonclassical’’ and operatic
styles of singing with respect to spectrographic, electroglotto-
graphic, and sub- and supraglottal pressure measurements and
arrived at a definition of ‘‘belting’’ on the basis of their findings:
‘‘Belting is a manner of loud singing that is characterized by
consistent use of ‘chest’ register (>50% closed phase of glottis)
in a range in which larynx elevation is necessary to match the
first formant with the second harmonic on open (high F1)
vowels, that is ∼G4-D5 in female voices.’’
Bestebreurtje and Schutte5 studied resonance strategies and
EGG contact quotients that their single subject used in belting
and an ‘‘unoptimized speechlike’’ style of singing in some
vowels sung on the pitches G4 and B4-flat. They found that
for /e/, the second formant F2 was close to the fifth harmonic
partial, whereas for /i/, F1 was midway between the first and
second partials. They concluded that the loud bright sound of
the belting style is achieved by the implementation of resonance
strategies that enhance higher harmonics. The contact quotient
for belting averaged above 52%.
In a third single subject study, Evans and Howard6 compared
the contact quotient in the belting and opera styles in some
vowels sung on the pitches C4, E4, C5, and E5. For the opera
style, the contact quotient varied in the range 18–34%, whereas
the belt varied in the range 43–57%. Some variation with F0
was observed.
The purpose of the present study was to identify characteris-
tic properties in some substyles of belting. Examples of the var-
ious belting substyles produced by a single subject were
recorded. To ensure that these examples were representative
of the respective substyles, a listening test was carried out by
a
It should be noted that Estill and also many other authors, including Schutte
and Miller4 and Evans and Howard,6 use the term ‘‘closed phase’’ instead of
contact phase.
Johan Sundberg, et al Belting Substyles: Phonatory and Resonatory Characteristics
FIGURE 1. Music of the song excerpt used in the recordings.
an expert panel. The examples were analyzed with regard to
subglottal pressure, voice source, and formant frequency char-
acteristics.
METHOD
The single subject of the study was coauthor L.P., who is a voice
coach and professional singer. She performed examples of five
substyles of belt commonly heard in musical theater: heavy,
brassy, ringy, nasal, and speechlike, and also classical voice
production for comparison.1
The song material was an excerpt taken from ‘‘Everything’s
Coming Up Roses’’ from Gypsy (1959, music by Jule Styne and
lyrics by Stephen Sondheim), Figure 1. She sang this example
first twice with the original lyrics and then twice with the sylla-
ble /pae/ replacing the syllables of the lyrics. Finally, she sang
diminuendos repeating the syllable /pae/ on the pitch of F#4
(370 Hz). The subject sang each of these examples in each of
the five substyles of belting and also, for comparison purposes,
in classical style of singing.
The recordings were made in the anechoic chamber of the
Linguistics Department of Stockholm University. Figure 2
shows the experimental setup in terms of a block diagram.
The audio signal was picked up by a Bru€el & Kjaer (Naerum,
Denmark) condenser microphone (B&K 4190 2435611) at a dis-
tance of 30 cm. A Gaeltec (Dunvegan, Isle of Skye, Scotland)
S7b pressure transducer, which the singer held in the corner of
her mouth, was used to capture the oral pressure. The oral pres-
sure during the occlusion for the consonant /p/ was used as an
estimate of subglottal pressure. The electroglottograph signal
was recorded from a Glottal Enterprises (Syracuse, NY) EG2
two-channel electroglottograph. These three signals were digi-
tized and recorded onto different channels of a wav-file using
SoundSwell software (Hitech Development, Solna, Sweden).
The audio signal was calibrated by sustaining a vowel, the
sound pressure level (SPL) of which was measured by means
of an Ono Sokki LA-210 (Ono Sokki Co., Ltd., Yokohama,
FIGURE 2. Experimental setup during the recordings in the an-
echoic chamber.
45
Japan) sound level meter held next to the recording micro-
phone. The SPL value thus observed was announced in the
recording. The pressure transducer was calibrated by first
immersing it into water at a depth that was measured and
announced in the recording and then holding it in free air.
As the experiment was run on one single subject only, it was
important to evaluate the examples of the various vocal sub-
styles recorded. This was done in terms of a listening test
with eight experts. The experts were all active as professional
teachers of singing in these styles and thus thoroughly
acquainted with the styles concerned. In the test, the stimuli
were divided into three groups, (1) the song excerpt with the
original lyrics, (2) the song excerpt sung on the syllable /pae/
and, (3) the four first (loudest) tones from the diminuendos on
the syllable /pae/, all produced in each of the six vocal styles.
The stimuli were copied onto three test files, one per stimulus
group. In each group, each stimulus occurred twice. The stimuli
were separated by 5-second long pauses and appeared in
random order. The total duration of the three test files was
460 seconds.
The test files were copied onto a CD and sent to each panel
member. Their task was to classify the stimuli on paper sheets.
Before the test, proper anchor stimuli were presented, typical of
the various vocal substyles (Table 1).
The voice source was analyzed by inverse filtering the audio
signal by means of the custom-made Decap program (Svante
Granqvist, KTH, Sweden), complemented by the derivative of
the EGG waveform (dEGG). Two criteria were applied for
tuning the inverse filters: (1) a ripple-free closed phase and (2)
synchronicity of the main positive peak of the dEGG and flow
waveform discontinuity representing the closing of the glottis.7
The frequencies and bandwidths of the inverse filter settings and
the resulting flow glottograms were saved to files together with
their derivatives and the dEGG signals. From the flow
TABLE 1.
List of the Anchor Stimuli Used in the Listening Test
Heavy: Lisa Kirk Big Time from Mack & Mabel, music and
lyrics by Jerry Herman
Brassy: Ethel Merman There’s No Business Like Show
Business, from Annie Get Your Gun, music and lyrics
by Irving Berlin
Ringy: Debbie Gravitte Secret Love from Calamity Jane,
by Sammy Fain and Paul Francis Webster
Nasal: Patti Lupone As Long As He Needs Me from
Oliver!, music and lyrics by Lionel Bart
Speechlike: Idina Menzel No Good Deed from Wicked,
music and lyrics by Stephen Schwartz
Classical: Beverly Sills Una Voce Poco Fa, from Il Barbiere
di Siviglia by Giacomo Rossini
46 Journal of Voice, Vol. 26, No. 1, 2012

repeated stimuli. The classifications given by listener 8 were

TABLE 2.
Consistency of Listeners’ Classifications
Rater
Stimulus Group 1 2 3 4 5 6 7 8
Single tones /pae/ 67 100 100 83 100 83 100 50
Song /pae/ 33 100 100 50 100 100 83 33
Song, lyrics 83 100 100 50 100 67 33 50
The numbers show, for each listener and for the indicated stimulus group,
the percentage of identical classifications of the six repeated stimuli.
glottogram were determined (1) period T0, (2) closed phase, (3)
pulse amplitude, (4) maximum flow declination rate (MFDR),
and (5) the level of the voice source fundamental relative to
that of the second source spectrum partial, H1-H2. The closed
quotient Qclosed was calculated as the ratio between the closed
phase duration and T0, and the normalized amplitude quotient
(NAQ) was calculated as the ratio between the pulse amplitude
and the product of MFDR and T0. Unfortunately, the speechlike
/pae/-diminuendo examples could not be inverse filtered
because of nasalization, since nasalization makes the transfer
function of the voice organ too complex.
RESULT
To check the consistency of the listening panel members, each
stimulus occurred twice, in a random order, in the test. The
numbers in Table 2 shows the consistency of listeners’ classifi-
cations in terms of the percentage of identical classifications of
discarded because they were far less consistent than those of
the remaining experts. The remaining listeners responded con-
sistently between 61% and 100% of the stimuli. This suggests
that their classifications represented rather reliable information.
The above results do not reveal how representative the vari-
ous stimuli were for the different styles. This, however, can be
seen in Table 3, showing the percentages of classifications that
were in agreement with the singer’s intentions. In most cases,
the singer’s intended styles resulted in high percentages. This
implies that most of the examples were representative.
As might be expected, the stimuli consisting of the four first
tones from the diminuendos on the syllable /pae/ seemed to
be somewhat more difficult to classify than the song stimuli.
Classical, speechlike, and heavy received the highest number
of ‘‘intended’’ classifications, whereas ringy, nasal, and brassy
received the lowest numbers. The four tone stimulus in the
Ringy style actually were classified in ways deviating from
the singer’s intentions in a majority of cases.
Summarizing, the listening test showed that most of the
stimuli were rather easy to classify and that in most cases, the
classifications made by the expert listeners were in agreement
with those of the singer. This supported the assumption that it
was meaningful to analyze in more detail the phonation and
resonance characteristics of these stimuli.
Voice source properties
Figure 3 shows subglottal pressure Psub and equivalent sound
level Leq, averaged across tones for the different vocal styles

TABLE 3.
Confusion Matrix Showing the Percentages of the Seven Experts’ Classifications That Were in Agreement With the Singer’s
Intentions in Each of the Stimulus Groups. Bold Digits Represent Classifications in Agreement With the Singer’s Intentions.
Classified as

Intended Brassy Classical Heavy Nasal Ringy Speechlike

Single pitches /pae/
Brassy 64 7 22 7
Classical 100
Heavy 7 93
Nasal 14 57 29
Ringy 64 36
Speechlike 14 86
Song /pae/
Brassy 93 7
Classical 93 7
Heavy 7 86 7
Nasal 64 7 29
Ringy 29 64 7
Speechlike 100
Song lyrics
Brassy 64 7 7 15 7
Classical 100
Heavy 7 93
Nasal 86 7 7
Ringy 14 7 65 14
Speechlike 7 93
Johan Sundberg, et al Belting Substyles: Phonatory and Resonatory Characteristics 47

FIGURE 3. Psub and Leq averaged across tones for the different vocal styles in the /pae/ version of the song excerpt and, in the left graph, also for
the diminuendo sequences.

in the /pae/ version of the song excerpt and in the diminuendo.
Because of technical problems, no Psub data were obtained for
the song example in classical. Heavy was produced with the
highest pressures and also showed the greatest variation of pres-
sure, as indicated by the high standard deviation in the song.
Figure 4A–E shows the MFDR, closed quotient, NAQ, H1-
H2, and pulse amplitude, respectively, for data observed in
the inverse filtering of the diminuendo sequences of the /pae/.
The large symbols included in the same graphs refer to values
observed in the /pae/ version of the song excerpt produced at
an F0 within ±1 semitone from the F0 used in the diminuendo
sequences of /pae/.
Heavy produced extreme values both in Psub and the various
glottogram parameters; MFDR and Qclosed were high, whereas
H1-H2 and NAQ were low. Classical assumed the opposite
extreme, in the sense that Psub, MFDR, and Qclosed were low
and H1-H2 and NAQ were high. Brassy showed high Qclosed
and low H1-H2.
For heavy and brassy, the song excerpt values are reasonably
close to the values observed from the diminuendo material. For
ringy, on the other hand, the song excerpt yielded values
quite separate from the diminuendo values. Speechlike assumed
a position with medium Psub, a low MFDR, H1-H2, and NAQ,
and high in Qclosed. In all these parameters except NAQ, speech-
like was close to brassy. The pulse amplitude showed no clear
differences between the vocal styles, neither for the /pae/-
diminuendo nor for the /pae/ version of the song.
Figure 5A–E summarizes the data shown in Figure 4 in terms
of the averages and standard deviations of MFDR, Qclosed,
H1-H2, NAQ, and pulse amplitude, respectively, as functions

FIGURE 4. A. MFDR, B. closed quotient, C. NAQ, D. H1-H2, and E. pulse amplitude for data observed flow glottogram of the diminuendo se-
quences of the /pae/. The large open symbols refer to values observed in the /pae/ version of the song excerpt and produced at an F0 within ±1 semi-
tone from the F0 used in the diminuendo sequences of /pae/.
48
FIGURE 5. Data shown in Figure 4 but here represented by their averages and standard deviations plotted as functions of the associated
averages of Psub.
of the average Psub for the diminuendo sequences of /pae/. As
expected, heavy assumes an extreme position not only with re-
gard to Psub but also to MFDR, H1-H2, and NAQ, whereas clas-
sical assumes the opposite extreme positions not only with
regard to a low Psub but also to low Qclosed and high H1-H2.
Brassy and ringy are close to each other with regard to Psub,
MFDR, NAQ, and pulse amplitude, whereas brassy as com-
pared with ringy is higher in Qclosed and lower in H1-H2. The
standard deviations are mostly small for classical and heavy.
The long-term average spectrum (LTAS) curves of the differ-
ent versions of the song are shown in Figure 6. Classical stands
out because of low levels between 500 and 3000 Hz, a marked
peak at 1600 Hz and a comparatively high level between 3000
FIGURE 6. LTAS curves of the different versions of the song excerpt.
Journal of Voice, Vol. 26, No. 1, 2012
and 4000 Hz. Also nasal shows a peak near 1600 Hz. All other
substyles show a marked minimum at 2500 Hz. The highest
peak appears near 350 Hz for classical. This is close to the
average F0 of 345 Hz of the song example; so this LTAS
peak simply reflects a strong fundamental. The other substyles
show a high peak near 750 Hz, that is, at twice the mean F0; so
in these styles, the second partial tended to be strong. Heavy,
ringy, and brassy have high levels between about 1000 and
1700 Hz, producing the highest LTAS levels. All styles except
classical show a marked peak near 3000 Hz. It should be kept in
mind, however, that the example was no more than about 10
seconds long and, disregarding the single, short, penultimate
note, had an ambitus of no more than a minor third.
The formant frequencies derived from inverse filter analysis
of the various syllables of the lyrics are shown in Figure 7. The
variation between styles appears to be unsystematic except for
F1 in classical, which is markedly lower than in the belting
substyles. The higher F1 values in the belting substyles would
reflect articulatory characteristics, such as a wider jaw opening,
a narrower pharynx, and/or spread lips.
The dotted curves in the Figure 7 show the frequencies of the
three lowest spectrum partials. The formant frequencies show
no consistent tendency to coincide with the two lowest partials.
The first formant is close to the second partial in no more than
two cases, in the vowel (e) in the word great, F0 z 343 Hz, and
on the vowel [ ] in the word the, F0 z 332 Hz, where the sec-
ond partial was about 15 and 10 dB stronger than the fundamen-
tal, respectively. The level difference between partials 1 and 2 in
the spectrum seemed to vary depending on F0 and vowel.
Johan Sundberg, et al Belting Substyles: Phonatory and Resonatory Characteristics
FIGURE 7. Formant frequencies, derived from the inverse filter analysis of the various syllables of the lyrics of the song excerpt.
DISCUSSION
Our investigation has focused on one single subject, which calls
into question the general validity of the findings. To counteract
this, we presented a listening test to an expert panel well ac-
quainted with the styles of singing concerned. The result of
this test revealed that all listeners except one mostly classified
the same stimulus the same way both the times it occurred in
the listening test. Thus, the classifications seemed to cause
rather limited difficulties to the panel. Moreover, the panel’s
classifications were mostly in good agreement with the inten-
tions of the singer subject; so it seems fair to conclude that
the examples analyzed were representative of the styles consid-
ered and hence merit detailed analysis. Furthermore, analyzing
one single subject implies that the same voice organ was used in
all styles thus entailing the method’s advantage that the subject
serves as her own control.
Schutte and Miller4 reported that the belt and legit styles of
singing differed in a number of ways. First, belt is produced
with an elevated position of the larynx ‘‘necessary to match
the first formant with the second harmonic on open (high F1)
vowels, that is, G4–D5 in female voices.’’ Recently, Bourne
and Garnier8 noted that F1, which they refer to as R1, was
higher in ‘‘chesty belt’’ than in legit and reported that
‘‘‘Chesty-belt’ quality’ was characterized by a tuning of R1 to
2f0, whereas ‘Legit’ did not show any particular resonance ad-
justment.’’ Our analyses concerned pitches in the range F4–Ab4
mainly, but there were several examples of high F1 vowels. As
shown in Figure 7, there was no tendency for F1 to be close to
the second spectrum partial. This observation is in accordance
with findings recently reported by Lebowitz and Baken.9 In-
deed, in the word gonn(a), the first formant was quite close to
the third partial. In the source spectrum, on the other hand,
the first partial was much stronger in classical than in the belt
styles, particularly the heavy.
We found no striking formant frequency differences between
the styles except that classical had a consistently lower F1 in all
vowels analyzed. As the examples sounded clearly different, as
quantitatively evidenced by the listening test, this result implies
49
that the voice source was the main factor distinguishing the
styles.
Belting has been found to be produced with high Psub10 and in
modal register.4 Our analyses of the voice source properties pro-
duced nicely stratified distributions of data parameters. On av-
erage across tones and pitches, Psub was highest in heavy and
lowest in classical, ringy, and brassy being similar to each other
in this parameter. Ringy, in particular, was quite close to classi-
cal. As belt is produced in modal register, the vocal folds can be
assumed to be thick, thus requiring high driving pressures.
A further support for this assumption is provided by Qclosed,
which was close to 0.5 in heavy and brassy and no more than
0.25 in classical. Ringy showed an intermediate Qclosed average
of about 0.3. The low quotient for classical may reflect a register
effect; in the F0 region concerned, classically trained female
singers rarely use a pure modal register.
The voice source fundamental was quite dominant in classi-
cal and much weaker in the other styles, heavy showing the
weakest, as might be expected from the Qclosed data. The pulse
amplitude average was rather similar for the different belting
substyles. This is not entirely surprising, given their Psub differ-
ences. An increase of Psub, a decrease of glottal adduction, and
a decrease of the closed phase tend to contribute to an increase
of the pulse amplitude. NAQ, inversely related to perceived de-
gree of pressedness,11 was low for heavy and higher and similar
for the other styles, including classical.
MFDR merits a special analysis because it reflects the vocal
tract excitation strength, that is, the contribution of the voice
source to the radiated sound level. Heavy showed the highest
MFDR values indicating that here the voice source was contrib-
uting substantially to vocal loudness. The means for brassy,
classical, and ringy were all similar, and the rank order was
the same as for the Leq.
We showed the results of the averages of the voice source pa-
rameters as function of Psub. This may be a promising method of
describing the voice use in a particular style of singing. The three
main control parameters of the voice source are Psub, F0, and glot-
tal adduction. The first two are easy to measure; but in the present
50
investigation, F0 varied within a very narrow range, no more than
a minor third, if a single low tone is disregarded. Therefore, F0
did not seem as an important parameter. Glottal adduction, by
contrast, was certainly a relevant source parameter, and Qclosed,
H1-H2, and NAQ separated the different styles more or less
clearly. At present, it seems difficult to decide which parameter
or combination of these parameters reflects glottal adduction
most faithfully. In future investigations of vocal styles, it may
be worthwhile to plot voice source data in a three-dimensional
graph with Psub, F0, and glottal adduction as dimensions.
Acknowledgments
Hassan Djamshidpey at Department of Linguistics of Stock-
holm University and Professor Sten Ternstr€ om kindly assisted
in the recordings. Financial support for this research and devel-
opment project was provided by University College of Music
Education in Stockholm (SMI), Sweden.
REFERENCES
1. Popeil L. Multiplicity of belting. J Sing. 2007;64:77–80.
2. Hollien H, Miles B. Whither belting? J Voice. 1990;4:64–70.
Journal of Voice, Vol. 26, No. 1, 2012
3. Estill J. Belting and classic voice quality: some physiological differences.
Med Probl Perform Art. 1988;3:37–43.
4. Schutte HK, Miller DG. Belting and pop, nonclassical approaches to the fe-
male middle voice: some preliminary considerations. J Voice. 1988;7:
142–150.
5. Bestebreurtje M, Schutte H. Resonance strategies for the belting
style: results of a single female subject study. J Voice. 2000;14:
194–204.
6. Evans M, Howard D. Larynx closed quotient in female belt and opera qual-
ities: a case study. J Voice. 1993;2:7–14.
7. Sundberg J, Thal
en M, Alku P, Vilkman E. Estimating perceived pho-
natory pressedness in singing from flow glottograms. J Voice. 2004;18:
56–62.
8. Bourne T, Garnier M. Physiological and acoustic characteristics of the fe-
male music theatre voice in ‘belt’ and ‘legit’ qualities. Paper presented at:
Proceedings of the International Symposium on Music Acoustics. August
25–31, 2010; Australia: 1–5.
9. Lebowitz A, Baken R. Correlates of the belt voice: a broader examination.
J Voice. 2011;25:159–165.
10. Hein M. Die Gesangtechnik des Beltings, eine Studie u€ber Atemdruck,
Lungenvolumen und Atembewegungen. Diss., Universit€at Hamburg;
2010.
11. Henrich N, d’Alessandro C, Doval B, Castellengo M. On the use of the
derivative of electroglottographic signals for characterization of nonpatho-
logical phonation. J Acoust Soc Am. 2004;115:1321–1332.