Journal of Voice

Vol. 9, No. 3, pp. 290-296

© 1995 Lippincon-Raven Publishers, Philadelphia

Measurement of Formant Frequencies and Bandwidths

in Singing

Harm K. Schutte, Donald G. Miller, and *Jan G. ~vec

Voice Research Laboratory, Department of Medical Physiology, University of Groningen, Groningen,
the Netherlands; and *Department of Experimental Physics, Palaclc¢ University, Olomouc, Czech Republic

Summary: That singers under certain circumstances adjust the articulation of

the vocal tract (formant tuning) to enhance acoustic output is both apparent
from measurements and understood in theory. The precise effect of a formant
on an approaching (retreating) harmonic as the latter varies in frequency during
actual singing, however, is difficult to isolate. In this study variations in am-
plitude of radiated sound components as well as supraglottal and subglottal
(esophageal) pressures accompanying the vibrato-related sweep of voice har-
monics were used as a basis for estimating the effective center frequencies and
bandwidths of the first and second formants. Key Words: Singing--Vocal tract
resonance--Formant tuning--Formants--Formant frequency--Formant band-
width--Vibrato.

One of the methods we have frequently used to the vocal tract. In such moments the standing wave
gain insight into the resonatory function in singing is effect is evident in the fact that the wave continues
that of measuring the variation in time of that which uninterrupted through the open phase of the glottal
we have generally called supraglottal pressure. The cycle, where it otherwise would be largely damped
concept is simple enough: A wide-band miniature in a poorly tuned, " u n r e s o n a n t " state.
pressure transducer mounted on a catheter passed Such strong standing waves in the vocal tract are
through the nares is positioned close to the glottis. also evident in the radiated sound, measured by a
Since the glottal end of the vocal tract forms a pres- microphone in front of the singer. B e c a u s e the
sure antinode, the space just above the glottis is the mouth opening has a stronger reflecting effect on
most favorable location for measuring the ampli- low-frequency sound emerging from the vocal tract
tude of standing waves in the vocal tract, at least for than on high frequencies, however, the microphone
the lower resonances. These amplitudes can be con- signal (and thus the sound we hear) gives a picture
siderable--we have measured peak-to-peak values of the pressure waves in the vocal tract that is
of 100 cm H 2 0 pressure in a soprano singing a high s o m e w h a t o b s c u r e d by the e m p h a s i z e d high-
note (1)---and often result in a waveform that shows frequency component (see, e.g., Fig. 9). Simulta-
a clear dominance of one or another resonance in neous registration of the two signals during singing
allows us to compare them, affording insight into
vocal tract resonances that underlie the radiated
sound.
Accepted July 15, 1994. The purpose of the present study is twofold: (a) to
Address correspondence and reprint requests to Dr. H. K.
Schutte at Voice Research Lab., Dpt. Med. Physiol., University show how audio and supraglottai pressure signals
of Groningen, BLOEMS1NGEL 10, NL 9743KZ, Groningen, can, under certain conditions, yield precise and re-
The Netherlands. liable information about formants in singing, and,
This article was presented at the Voice Foundation, 22rid An-
nual Symposium: Care of the Professional Voice, Philadelphia, by extension, (b) to indicate how these signals can
PA, U.S.A., June 1993. serve as informal feedback for locating and adjust-

290
 FORMANT MEASUREMENTS IN SINGING 291

ing the formants for improved "resonance" in sing- not expect the F 1 measurements to be as well de-
ing. fined as those for F 2.

METHOD RESULTS AND DISCUSSION

With use of wide-band pressure transducers F 2 dominance
mounted on a flexible catheter 1.6 mm in diameter Figures I-4 depict details of phonations illustrat-
(type Honeywell HD56, size F6), it is possible to ing moments in which resonance from the second
obtain pressure measurements from a point in the formant dominates the vibrato-based modulation of
pharynx just above the glottis (supraglottal pres- the SPL. Figure 1 is taken from a series of vowels
sure) with a minimum of invasiveness during sing- on a sustained G4 sung by the tenor subject, and
ing. The subject feeds the catheter through the nose Fig. 3 from the final note of an ascending scale seg-
and "swallows" the distal end. The catheter is then ment on the vowel /a/ ("open oh") by the bass-
adjusted and fixed at a point where the two lower baritone. Figures 2 and 4 present osciliograms of
transducers are in the esophagus and the remaining the microphone (audio) and supraglottal pressure
(proximal) one is in the desired supraglottai posi- (supra) signals at points (P, Q, R, S) indicated on
tion. The signals from all three transducers, along Figs. 1 and 3, respectively.
with signals from an electroglottograph (Fourcin Figure 1 shows five cycles of vibrato in the top
Laryngograph) and a microphone 30 cm in front of panel. Below it are the corresponding SPL modula-
the subject, are registered by an instrumentation tions of three signals: unfiltered audio, as well as
recorder (Teac XR-510). For later processing of the audio and supraglottal pressure (supra) signals from
signals, curves were written out on an AstroMed
MT 9500 thermal array recorder and analyzed with 400 fund. rreguenc.Y . . . . . . . . . . . . . . . . . . . . . . . . .

the CSL 4300 (Kay Elemetrics) analysis package. ~,~ 390

Our subjects were an operatic tenor (lirico- o 380
spinto), making an international career, and a bass- 370

baritone (one of the authors) with many years of 81 audio.

............. !...'.!. . . . . . . . . . . . . . .
" "
experience singing professionally in opera and con-
cert. They recorded scale passages and sustained
notes on various vowels in the untreated acoustic
environment of the laboratory.
The signals displayed in the first four figures were 66 .

selected for their capacity to reveal dimensions 63 . . . . . . . . . . . . . . . . . . . . . . . . . . i...i. . . . . . . . . . . . . . . . . .

(center frequency and bandwidth) of a dominant supra, 1 1 -- i" '!
resonance in a sung phonation. Specifically, a mo-
ment of high sound pressure level (SPL) occurs
when one of the harmonics of the voice source
sweeps across the resonance of interest during the
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 I.O
frequency modulation of vibrato. Further, in the
Time ( s e c )
phonations considered here, the location of the har-
monic, with respect to the resonance, is such that FIG. 1. Segment of sustained/i/on the pitch G4 by tenor sub-
ject. Top: Fundamental frequency as it modulates with vibrato.
the frequency sweep includes and exceeds the en- (The somewhat jagged aspect of the frequency contour is due to
tire - 3 - d B bandwidth. a limited sampling frequency of the signal.) Middle: SPL of the
While it is not unusual for the second formant total signal from a microphone 30 cm in front of the subject, as
well as of that signal through a bandpass filter that isolates the
(generally >1,000 Hz) to be completely swept in fourth harmonic (filter characteristics: order 100, 1.4-1.7 kHz
vibrato by one of the harmonics, the first formant, bandpass Blackman filter). Bottom: SPL of the signal from a
which seldom exceeds 700 Hz in "classical" voice wide-band pressure transducer placed just above the glottis.
Solid vertical lines drawn through SPL peaks show that these
production by male singers, presents a different coincide and occur at recurrent, similar points in the Fo modu-
case. A generous vibrato modulation amplitude of lation in the ascending and descending phases (points P and R).
10% (nearly a full tone) will produce a sweep of 70 The dotted vertical lines are drawn through points on the filtered
audio signal 3 dB below peak SPL. Their projections on the Fo
Hz at the frequency of the formant, hardly enough curve give the limits of the ( - 3 - d B ) bandwidth. SPL values, in
to cover adequately the bandwidth. Thus, we can- this case, have arbitrary reference value.

Journal of Voice, Vol. 9, No. 3, 1995

292 H. K. SCHUTTE ET AL.

Q - F o max R - SPL max S - Fo rain

FIG. 2. Oscillograms of microphone and (un-

calibrated) supraglottal p r e s s u r e , taken from
points marked with letters (Q, R, S) in signals of
Fig. I. Each token of - 2 . 5 cycles is adjusted so
that m o m e n t s of glottal closing (cl) are aligned.
Note the high-frequency e m p h a s i s of supra with
t'l 4 respect to audio.

supra ,~,[ ita

el el cl

a bandpass filter designed to isolate the fourth har- In Fig. 2 the oscillograms of selected points in the
monic. The fact that SPL modulation occurs at vibrato cycle are adjusted in time by aligning the
twice the rate of Fo modulation is clearly a result of audio and supra signals at the moment of glottal
the sweep of H4 across F 2 in both directions. The closing ("cl"). Both the maximal amplitude of the
projection of the SPL peaks on the frequency curve wave (within a larger time segment) and its domi-
(solid vertical lines) shows a difference in formant nant four-part structure allow us, by visual inspec-
frequency of 8 Hz between the ascending and de- tion, to locate the point of SPLmax within a narrow
scending phases (points P and Q). The projection of time range. Supraglottal pressure is given here with-
the -3-dB level of filtered F2 (the other SPL curves out amplitude calibration, essentially as a micro-
differ only marginally) on the frequency gives an phone signal.
effective bandwidth of 16 Hz at fundamental fre- Figure 3 shows 2.5 cycles (note the different time
quency. Since the sweep of the resonance takes scale) of vibrato in the top panel and similar SPL
place at the fourth harmonic, however, F 0 must be signals to those of Fig. 1, presenting five sweeps
multiplied by 4 to calculate F 2 (1,572 ascending and past a dominant F2. As with the tenor, the center
1,540 descending) and its bandwidth (64 Hz). frequency of the formant is higher in the ascending
phase. Since in this phonation it is the third har-
, i , ,
360 ~ . . . . 0 monic that sweeps the formant (see the oscillo-
"~ 350- ~s.= grams in Fig. 4), the Fo values must be multiplied by
3: F 2 is 1,035 Hz ascending and 1,005 Hz descend-
o 330
320 frecluency
f~equency vv {I.~ [ ~ z -]
• i
ing; bandwidth is 45 Hz.
81 audio, u n f i l t e r e d i - The extent to which the vibrato modulation
sweeps the bandwidth has important implications
~ 75
~ 72 for the accuracy of measurement of formant quali-
~ 69 ties. The double sweep in the F 2 signals presented
~ a6
63 here affords the determination, with a high degree
60
57 nu(ho fllteied F
of precision, of the effective formant frequency; a
direct measurement of the less precise parameter,
bandwidth, subject to reservations noted; and clear
~" 69 evidence of the correlation of the tuned moment in
~ 66
~ 63 the vocal tract with the moment of high SPL in the
60 radiated sound.
57 ...... - . . . . . . . . . . . .
i i i i i As we have argued elsewhere (2), a complete de-
0.0 0.1 0.2 0.3 0.4 0.5
scription of the modulation of sound in singers' vi-
Time (sec)
brato must consider not only the varying proximi-
FIG. 3. Signals similar to those of Fig. i. This time the s e g m e n t ties of harmonics to formants, but also vibrato-
is the vowel IOI on the pitch F4 s u n g by a b a s s - b a r i t o n e subject.
Note the difference from Fig. 1 in time scale (filter characteris- based modulations of the voice source spectrum.
tics: order 100, 0.9-1.2 k H z b a n d p a s s Blackman filter). The difficulty of distinguishing between source fac-

Journal of Voice, I/ol. 9, No. 3. 1995

 F O R M A N T M E A S UREMENTS I N S I N G I N G 293

8 - F o min R - S P L max Q - F o max

audioF ~
FIG. 4. Oscillograms, similar to those of Fig. 2.
taken from points marked with letters (S, R, Q)
in Fig. 3. cl, glottal closing.

supra ~,~/"~ r,.,./""/'~,.,,.,,,/

cl ol cl

tots and vocal tract (resonance) effects makes any indicated dominance of F, is confirmed in the os-
analysis of vibrato complex. [For a theoretical over- cillograms (Fig. 6).
view of the problem, see Rothenberg et al. (3)]. A If we make the plausible assumption that F, is
list of modulating factors should include for voice reached at the peak of the first vibrato cycle, it is
source: intensity (with varying subglottal pressure) possible to measure a half-bandwidth. Extrapolat-
and source spectrum (with varying closing speed ing from the 12 Hz measured (24 Hz for the full
and closed/open quotient); and for vocal tract: width x 2 for H E) gives 48 Hz, also a plausible
physical movement in vibrato; and varying band- value. However, there are problems in taking such
width with closed quotient. an approach.
These have a relatively small effect in the case of In the case of the signals with F, dominance, the
our F2 examples. Rothenberg et al. (3) found that
the greatest variations in the voice source are cor- 360
550 p
related with the maxima and minima of frequency 540 ' ' ' ' ' ' ' ~ '
0
modulation, while the resonant points in our F 2 ex- ~-
v
330
o 520 ;~., .
amples occur near the center of the cycle. The vi- -- 510

brato-based variation in mean subglottal pressure 300

290
(estimated from esophageal pressure) does not ex-
ceed 3% of the mean pressure, and we disregard the 78

still narrower fluctuations between points P and R

in Fig. 3. The difference between the formant fre- g 72
quencies in the ascending and descending phases of audio, unfiltered

69 ~
vibrato is a consistent feature of both F 2 examples
given here, and it witnesses, incidentally, to the
precision of the measurements. We attribute this ~ 57
difference to vibrato-based movement of the vocal g
tract, for example, vertical oscillation of the larynx.
u~ 5460 a:: ~ ~ i audio,filtereFd,
Bandwidth measurements varied within a narrow 51
range and showed a not more than marginal differ-
ence between ascending and descending phases.
g
~ 60
F l dominance
57 . . , i , r t
Figure 5 is composed similarly to Figs. 1 and 3. 0.0 0.! 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 ,.0
This time the signals come from two pitches (three Time (sec)
vibrato cycles each) of an ascending scale sung by
FIG. 5. Signals similar to Fig. I, giving the segment E~,4-F4 of a
the tenor. The filter is chosen, however, to isolate scale on the vowel/a/sung by the tenor subject. Each of the two
the harmonic (H 2) that engages the dominant F I res- pitches gets three cycles of vibrato. Here the filter is designed to
select the dominant second harmonic, which is resonated by F I.
onance. There is no complete sweep of the formant,
See text for discussion of formant frequency and bandwidth (fil-
and the SPL curves, at least at the beginning of the ter characteristics: order 100, 0.5-0.8 kHz bandpass Blackman
segment shown, closely follow the F o curve. The filter).

Journal of Voice, Vol. 9, No. 3, 1995

294 H. K. S C H U T T E E T A L .

P- F o max Q - F o mb filters of 200-900 and 900-2,200 Hz, chosen to dis-

play the whole-frequency ranges of F I and F 2, re-
spectively, rather than simply the (dominant) har-
monic of interest. Comparing the curves for audio
audio
(Fig. 7) and Psupra (Fig. 8), we find a particularly
close correspondence between these two at mo-
ments where a single formant is dominant, i.e., to
the left of the vertical line (E [4-F4) in the frequency
component 200-900 Hz and to the right (G4-A4) in
supra
the frequency component 900-2,200 Hz. Note,
however, the discrepancy on the higher notes be-
ol ol tween unfiltered Psupra, which closely follows supra
FIG. 6. Oscillograms, similar to those of Fig. 2, taken from the 900-2,200 Hz, and unfiltered audio. Unfiltered au-
points marked with letters (P, Q) in Fig. 5. cl, glottal closing. dio misses the sharp valleys of supra, as these are
filled in by a strong singer's formant component
argument for neglecting the variation in the voice that appears at F 0 max points, evident in the oscillo-
source is much weaker. The point of Fo max in vi- grams (particularly audio) of Fig. 9.
brato appears to be the closest approximation of the Further observations
second harmonic to the first formant. Since subglot- Even if we are unable to determine precisely the
tal pressure and the glottal closed quotient also vary value o f F I on the El, and F in the scale passage, our
with F 0 in vibrato (3), however, the contribution of
TENOR, Eb4-A4, /a/ AUDIO
these source factors to the modulation of SPL can- 450 ' ~ '
not be easily dismissed. Furthermore, since there is
no indication that the second harmonic passes F~, it 400

is not obvious that the center frequency of F~ has I

550
been reached. Indeed, in the frequency range -300
Hz (-D4, marginally higher for tenors), in male 300

singing voices one can observe that if F~ dominance

is established, there is generally a tendency to ad- 8O
just the formant upward to stay on the upper margin "o
~ 70
of the excursion of the second harmonic (4). A sim-
ilar relationship of F t and H 2 characterizes the 60
higher pitches in a female singing voice production 80
called "belting," although vibrato is generally not
~" 70
an important factor there (5).
~ 6o
Transition of dominance
The two pitches depicted in Fig. 5 are part of an 50
80
ascending scale. By examining the extension of that
scale, we discover what happens when the voice 70
"o
passes the (pitch) barrier of Ft dominance. In Figs.
7-9 we see that the scale step F4-G4 is character- ~- 60
ized by a decisive change in dominance from F I - H 2
50
to F 2 - H 3. The overall SPL stays fairly constant, but 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
the resonance responsible for the dominant part of Time (sec)
the sounds shifts. Remarkably, the F and G, located FIG. 7. Signals from the scale segment shown in Fig. 5, ex-
on opposite sides of the point of transition, indi- tended by three vibrato cycles of G4 and one of A4. The two
bandpass filtered audio SPL signals are intended to include the
cated by a solid vertical line, are perceptually well whole range of the first and second formants, respectively. The
matched, illustrating the skill of the singer in dis- solid vertical line is located at the border between pitches F4 and
guising and minimizing the change. G4, where the major source of SPL shifts abruptly from Fj to F2.
" C h e s t " and " h e a d " are terms that some singers have used to
In Figs. 7 and 8 we have replaced the narrow- designate this shift. The dotted vertical line shows the coinci-
band filters of Figs. 1, 3, and 5 with wider-band dence of vibrato Fo mi. with high SPL in the F 2 range.

Journal of Voice, Vol. 9, No. 3, 1995

 FORMANT MEAS UREMENTS IN SINGING 295

TENOR, Eb4-A4, / a / SUPRA velop names for the difference in feel (and sound)
450
between the different dominances. Our figure gives
400 two terms, "chest" and "head," which have some-
I
times been applied to this transition.
350
o Comparing the various signals for the identical
500 phonation gives rise to further general observa-
tions. The effective high-frequency emphasis in the
80 audio signal is the most characteristic feature of the
difference between audio and supra oscillograms.
v 70
This is most evident in the F o max of Fig. 9, where a
60 strong singer's formant (at the eighth harmonic)
80 dominates the audio, appearing much weaker in su-
pra. The Psupra signal, on the other hand, is presum-
~" 70
"a
v ably free from the influence of the room acoustics,
g 60 which may be a significant factor in audio SPL.
Comparison of various signals taken simulta-
50
80 neously brings us-to the heart of the matter regard-
ing the extraction of information from signals that
~"
"o
70 elude precise quantification, such as those pertain-
g- 6o
ing to F~ in this study. Fi dominance, especially
i/i
supra 900-2200 Hz
where it resonates the second harmonic, is a con-
50 i i I i t spicuous feature of a sung phonation and can be
0.0 0.2 0.4 0.6 0.8 1.0 ~.2 1.4 1.6
easily followed in real time by monitoring oscillo-
Time (sec)
grams, as well as spectrum analysis. Feedback is
FIG. 8. The same scale segment as Fig. 7, this time with the SPL
signals taken from the supraglottal pressure. The filtered signals salient, and one can learn to associate the perceived
follow similar curves to those characterizing audio. The portion complex "signal" (auditory and proprioceptive)
of unfiltered Psup,a to the right of the vertical line, however, with the characteristic oscillogram. Careful com-
differs from unfiltered audio in that it effectively reduces the
prominent contribution of the singer's formant at the peaks of the parison of signals taken simultaneously leads to the
vibrato cycle. ability to predict one signal on the basis of another
and even to predicting signals on the basis of per-
ception. Gradually a coherent picture of the various
data leave little doubt that H 2 passes F~ in the move parameters emerges.
from F to G (marked by the solid vertical line).
Typical of tones characterized by F r H 2 dominance CONCLUSIONS
is the fact that the harmonic does not pass the for- This study, more narrowly defined, leads us to
mant in vibrato, but approaches it as an upper limit. the following conclusions:
In Figs. 1 and 3, the magnitude of the shift of F 2
p - F o max Q - F 2 SPL max
frequency between points P and Q measured is on
the order of 2.5%, even near the center of the vi-
brato sweep. The full extent of the (peak-to-peak)
formant frequency shift in vibrato is likely to be audio
considerably greater; the method presented here,
however, does not allow us to estimate this accu-
rately.
In Figs. 7 and 8, although F~ appears to be
reached already at the top of the vibrato cycle on supra " ~ , ~
E~4, the whole of the pitch F4 remains under F~
dominance. This scale segment is presented as a cl c!

particularly vivid illustration of the move from F~ FIG. 9. Oscillograms similar to those of Fig. 2, taken from points
dominance to F2 dominance. Other comparable marked with letters in Figs. 7 and 8. The signals show the evident
dominance of the third harmonic at F2 max (indicated with B), and
scale segments may show it less conspicuously or the strong singer's formant on the eighth harmonic is conspicu-
not at all, but it is not surprising that singers de- ous in the microphone signal at A. cl, glottal closing.

Journal of Voice, Vol. 9, No. 3, 1995

296 H. K. SCHUTTE ET AL.

1. I n the p r e s e n c e o f v i b r a t o , m i c r o p h o n e a n d Jan ~;vec is on research stay in the Voice Research Lab.

supraglottai pressure signals can sometimes pre- We appreciate the assistance of Meindert Goslinga for his
work on the figures.
c i s e l y r e v e a l c e n t e r f r e q u e n c i e s a n d , less p r e c i s e l y ,
b a n d w i d t h s o f d o m i n a n t f o r m a n t s in sung p h o n a -
tions. REFERENCES
2. T h e s a m e signals c a n a l s o p r o v i d e c o n s p i c u -
1. Schutte HK, Miller DG. The effect of F0/FI coincidence in
o u s e v i d e n c e o f c l o s e p r o x i m i t y o f h a r m o n i c s to soprano high notes on pressure at the glottis. J Phon 1986;
d o m i n a n t f o r m a n t s , e v e n in c a s e s w h e r e t h e for- 14:385-92.
m a n t c e n t e r f r e q u e n c y c a n n o t b e d e t e r m i n e d un- 2. Schutte HK, Miller DG. Acoustic details of vibrato cycle in
tenor high notes. J Voice 1990;5:217-23.
equivocably. 3. Rothenberg M, Miller DG, Molitor R. Aerodynamic inves-
3. T h e f o r m a n t f r e q u e n c y e x h i b i t s a c o n s i d e r a b l e tigation of sources of vibrato. Folia Phoniatr 1988;40:244-
m o d u l a t i o n , c o r r e s p o n d i n g with v a r i a t i o n in v o c a l 60.
tract during vibrato. 4. Miller DG, Schutte HK. Towards a definition of male 'head
register', passaggio, and 'cover' in Western operatic sing-
ing. Folia Phoniatr Logop 1994;46:157-70.
Acknowledgment: This work was supported in part by 5. Schutte HK, Miller DG. Belting and pop, nonclassical ap-
grants from the Jan Patoeka Foundation and Hlfivka proaches to the female middle voice: some preliminary con-
Foundation and a loan from the Civic Forum Foundation. siderations. J Voice 1993;7:142-50.

Journal of Voice, Vol. 9, No. 3, 1995