J. Acoust. Soc. Jpn.

(E) 7, 6 (1986)

Singer's high formant associated with different

larynx position in styles of singing*

Shigian Wang**•õ

**Physics Department of Faculty of Science and Music Department of

Faculty of Fine Arts, York University,

4700 Keele Street, Downsview, Ontario M3J 1P3, Canada

(Received 27 June 1985)

This paper is a further presentation and discussion of the author's voice research and
previous paper and film on singer's (high) formant associated with different larynx posi-
tion in styles of singing. The main results and points are : 1. The singer's high formant
pattern varies with similar characteristic (two to four high and often compact energy
peaks, including a possible extra peak fe, in the bright frequency range); 2. The fe is also
found in high laryngeal singing and does not necessarily appear as a typical acoustic
feature in all professional singing voice; 3. Sundberg's lowered larynx interpretation does
not explain the extra formant, singer's formant, and voice quality; 4. Sundberg's different
definitions of singer's formant are not inclusive of the various formant patterns of singing
associated with different laryngeal position and different types ; 5. The author also sug-
gests a definition for the singer's formant; 6. The low laryngeal position could be yet
regarded as a distinctive feature for the covered singing voice though not for the trained
singing voice; 7. The source characteristics and the interaction between the source and
the supra/sub glottal systems are possibly different in styles of singing associated with
various laryngeal positions.

PACS number : 43. 70. -h, 43. 75. - z

no singer's formant. He explained that the lowered

1. INTRODUCTION larynx during singing is the source of this singer's
The singer's (high) formant have been studied by formant.
Bartholomew,1) Rzhevkin,2) Vennard,3) Sundberg,4) Sundberg4,5) further defined the singing formant
and dozens of other researchers. as an extra formant in the formants of the singing
Bartholomew1) found and defined the singer's voice. That is, the singing formant is an extra high
(high) formant (hereinafter singer's formant) as a spectrum envelope peak near 2.8 kHz located be-
prominent formant which averages around 2.8 N 2.9 tween the third (F3) and fourth (F4) formants of vowel
kHz for male voice, and 3.2 kHz for female voice. sounds produced in male Western opera and concert
He also claimed that the coloraturae sometimes have singing. Sundberg has different meanings for the
concept of the singing formant, and we will refer to
* This paper was invited to be presented at the Confer- and discuss those meanings later in this paper. To
ence, and originally submitted to the Transactions, of distinguish this from his other meanings and for the
the Committee on Speech Research, Acoustical Socie- convenience of this discussion the term "extra for-
ty of Japan 1984 (Tokyo). mant/peak" and its symbol fe will primarily be used
† Present address : Speech Communication Laboratory,

the Institute of Acoustics, Chinese Academy of Sci- hereafter instead of Sundberg's term "singing for-
ence, 5 Zhongguancun Road, Beijing, The People's mant."
Republic of China Sundberg4) assumed that lowering the larynx

303
 J. Acoust. Soc. Jpn. (E) 7, 6 (1986)
"explains*
the singing formant and other formant damage) and for the listener (because of poor quality
frequency differences between normal speech and of sound produced by some singers) (cf. Refs. 7 N 13),
male professional singing." That is5) : "The acoustics etc.).
of the vocal tract when the larynx is lowered are This author", ") investigated tenors singing in
compatible with the generation of just such an extra different styles [some Chinese, early music and
formant if the area of the outlet of the larynx into Western opera (here, i.e., covered voice)] and his
the pharynx is less than a sixth of the area of the results indicate that revisions to the above concepts,
cross section of the pharynx, then the larynx is which support some of his previous viewpoints,")
acoustically mismatched with the rest of the vocal should be made as follows :
tract," and it5) "acts as a separate resonator capable 1. Laryngeal height is not a reliable way to dis-
of adding an extra formant to the transform function tinguish a trained singer from an untrained one. Not
of the vocal tract."5) " ... it has a resonance frequency all trained singers maintain the low laryngeal posi-
of its own, largely independent of the remainder of the tion.
tract. The one-six condition is likely to be met when 2. In both high and low laryngeal styles of sing-
the larynx is lowered. ... The lowering of the larynx, ing, some trained singers are capable of producing
in other words, seems to explain the singing formant bright healthy voice and its associated acoustic fea-
peak. ... The result is to shift downward all formant tures. Lowering of the larynx therefore does not
frequencies other than the larynx-dependent extra explain the production of singer's formant and
formant. ... The lowering of the larynx, then, explains brightness. Singing with a high larynx in and of
not only the singing-formant peak but also major itself does not necessarily produce a voice of poor
differences in the quality of vowels in speech and quality or cause poor vocal health.
singing." Sundberg5 claimed that this extra for- This author and others14,15,17,18)also found that
mant "is typical of all voiced sounds sung by profes- the amplitude and frequency of singer's formant are
sional male singers. Indeed, its presence, regardless varied with pitch, vowel and intensity. Recently
of the pitch, the particular vowel and the dynamic Sundberg (with Johnson and Wilbrand)19) reported
level has come to be considered a criterion of quality; in publication that some sopranos raised their laryn-
the extra peak has been designated the singing for- ges during singing. However it was noticed that
mant." Sundberg4) also extended Bartholomew's1,6) Sundberg's concept and his explanation of lowering
conclusion that in the soprano voice "No 'singing the larynx for singer's formant were limited to male
formant' is likely to be produced." Western opera and concert singing voice.
Larynx location during singing phonation has been
regarded by previous researchers not only as being 2. QUESTIONS
particularly important in producing the singing for- In this author's research (1980 - 1983) acoustic
mant and to affect the singer's other formants, but questions arose concerning the differences between
also to affect the voice quality and vocal health of the the singer's larynx positions in different styles of
performer. Shipp and Izdebski7) found that non- male singing.
singers positioned the larynx upward in increased 1. What are the different and common features
voice frequency and tended to maintain a level at or of the radiated spectra between high and low laryn-
above the physiological resting position, whereas geal singing voices.
trained singers maintained laryngeal position below 2. What are the different and common features
the physiological resting position throughout their of the radiated spectra between speaking and (high
vocal range. and low laryngeal) singing voice.
In accordance with these findings and explanations 3. How do we comment on Sundberg's concept
another traditional concept has apparently been that of singer's formant which was interpreted as due to
singing with a high larynx is generally seen as "sing- lowering of the larynx and how do we explain the
ing in the pain" both for the singer (because of its acoustic features associated with high/low laryngeal
frequent association with vocal deterioration or singing voices.
Those questions should be answered in this re-
* In this paper all italicizations in quoted passages have search.
been added by the present author.

304
 S. WANG: SINGER'S HIGH FORMANTS

obtained from the corresponding cepstrum. The

3. METHOD sampling rate was 10 k/s, and the cutoff frequency
Ten trained tenor voices representive of the is 5 kHz. The predictor order was 14. A 25.6 ms
Chinese, early music and the Western opera singing segment of voice preemphasized at 6 dB/oct was
style were chosen for this study each with at least selected visually, multiplied by a Hamming window,
fifteen years of training and experience and in ap- and converted into both a magnitude spectrum using
parent good vocal health. Most of the subjects are the DFT and an idealized spectrum using the all-pole
soloists and some are highly acclaimed nationwide or model LPC.
regionally (Although these singers may sing in more
4. RESULTS
than one style, and each style, such as the Chinese
singing, has many kinds of singing styles, their voices This paper will deal only with the results of the
are here only labeled as exhibiting, but not generaliz- perceptual judgement of the timbre and the measure-
ing, either of the three styles). The study consisted of: ment of the larynx position which are related to the
(1) Perceptual judgement of timbre. following acoustic results and discussion.
(2) Measurement, during phonation, of the vertical 1) The voice of all subjects was rated as being in
distance of the larynx from its physiological the bright range although, as might be expected,
resting position. there were different degrees of brightness (see Table 1
(3) Measurement of acoustic features, including rating the brightness of sung vowels).
the first five formants, F1, F2, F3, F4, F5, the 2) The vowel /i/ and /A/ were judged as being
extra formant fe, and the fundamental fre- significantly brighter than /u/ (significance level p<
quency F0. 0.001).
Perceptual judgements were made by asking each 3) In the Chinese and early music style singers,
subject to sing (in full voice) three vowels : IA/, /u/. the larynges generally remained above the physiologi-
The sounds produced were rated as to professional cal resting position, especially in the high range of the
brightness on a scale of 1 to 7 by 19 voice teachers voice. The larynx of each subject singing in the
and singers at a meeting of the Boston Chapter of the Western operatic (covered) style was usually below
(US) National Association of Teachers of Singing. the resting position. This finding was very clearly
Vertical movement of the thyroid cartilage away demonstrated by two subjects who could sing in the
from its resting position was recorded and measured Western operatic, and either the Chinese or the early
by the use of photographic slides and videotapes, music style (see Fig. 1). While the height of the larynx
while the voice was simultaneously recorded on audio varied from singer to singer, the position of the
tape during phonation. The measurement was facili- larynx was highest for the vowel /i/, nearly as high
tated by placing a narrow strip of colored adhesive for /A/, and the lowest for /u/ in all styles of singing.
tape on the skin of the neck, at a right angle to the 4) In the Chinese and early music singers the
airway, overlying the tip of the thyroid prominence height of the larynx increased as the pitch went up.
in the physiological resting position. The extent of By way of contrast, the Western opera singers'
the vertical movement as measured on the slides and larynges moved progressively lower as pitch went up.
videotapes was then multiplied by a coefficient ob- These findings held roughly true for pitch up to about
tained by measuring a suitable facial landmark and G4, or 392 Hz (see Fig. 1).
comparing it to the measurement on the slide or 5) Two to four high energy peaks in some cases
videotape. including the extra peak fe between or near F3, or F4,
Acoustic features were determined by the spectral
analysis program on a PDP-11 computer with a
Table 1 Rating the brightness of sung
UNIX operating system at the Speech Communica-
vowels.
tion Laboratory of MIT, USA. This program was
designed to digitize a series of waveform files and
perform several kinds of spectral analysis. The
formants were determined by using both the Discrete
Fourier Transform (DFT) and Linear Predictive
Coding (LPC), and the fundamental frequency was

305
 J. Acoust. Soc. Jpn. (E)7, 6 (1986)

Fig. 2 A Chinese singing voice. vowel : /i/,

pitch: #C4 (277 Hz), LPC spectrum.

Fig. 3 A Western opera voice. sung vowel :

/i/,pitch:#C4(277  Hz),  LPC  spectrum.

Fig. 1 Measures of superior or inferior

vertical larynx position (VLP) with respect

to the physiologic rest position (zero line)

of four singers producing sustained phona-

tion at selected fundamental frequencies

(Fo). A Subject A (early music singing

voice), •› Subject B (Chinese singing
voice), •œ Subject B (Western opera sing-

ing voice), O Subject C (early music sing-

ing voice), •Ÿ Subject C (Western opera

singing voice), 0 Subject D (Western

opera singing voice).

Fig. 4 An early music singing voice. vowel:

were found in the spectra of not only the Western /i/,pitch:1)4(294Hz),  LPC  spectrum.

operatic voices, but also the Chinese and early music

singing voices in this study. Those peaks were found bandwidth between the centre frequencies of those
roughly between 1.8 and 3.8 kHz, which we shall peaks is roughly less than 1.5 kHz. The peaks were
call the Bright timbre Frequency Range, or BFR (see generally compact and higher in amplitude than the
Figs. 2 N 4). For each individual, the maximum high formants in the speaking voice. Usually F5 was

306
 S. WANG : SINGER'S HIGH FORMANTS

not close to those peaks.

6) In both high and low laryngeal singing, the
relative amplitude of the formants in the BFR in-
creased as the fundamental frequency increased.
7) Generally there were more formant peaks with
stonger relative amplitude in the BFR and weaker
F1 on /i/ spectrum than on /u/, with /A/ lying between
them (see Figs. 3, 5, and 6), especially for the same
subject and singing style.
8) All the formants frequencies were higher for
the Chinese and early music voices when compared
to the Western opera voice in the experimental range
(C4-G5) with three samples per half pitch step. Fig. 7 - - - A Western operatic voice.
These measurements were taken on two subjects, A Chinese singing voice. vowel: /i/, pitch:
each singing in two styles, and generally hold true #C4(277Hz),  LPC  spectrum.

Fig. 5 An early music singing voice. sung

vowel: /u/, pitch: C4 (262 Hz), LPC spec-
trum.

Fig. 8 Comparisons of singer's formants in

different styles. vowel: /A/, W1: Western
operatic voice of subject 1, CI.: Chinese
Fig. 6 A Western operatic voice. sung singing voice of subject 1, W2: Western
vowel : /A/, pitch: F4 (349 Hz), LPC spec- operatic voice of subject 2, E2: early
trum. music voice of subject 2.

307
 J. Acoust. Soc. Jpn. (E) 7, 6 (1986)

except for F1 in the vowel /i/ in which a slight inverse be different from them. In this research I also analyze
change occurred (see Figs. 7 and 8). some spectra of untrained voices produced with a
9) For the same subject and same vowel pho- high larynx. The amplitude of their high formants
nation most formants of low laryngeal (the Western are low or unstable. Further determination of the
operatic) singing voices, except for some cases, were physiological and acoustic differences between good
generally lower than those of the speaking voice. and bad singing with a high larynx would be an
However in high laryngeal singing (the Chinese and interesting topic for future research; the results could
early music singing) voice the formants roughly have important implications for the training and care
tended to shift upward, except F4 which was usually of the professional voice.
compacted to F3 and fe. The shifting direction of the Because the singing voices were rated as bright and
formants in high laryngeal singing is not as consistent we also found the extra formant fe in high larynx
as that found in low laryngeal singing. singing, the explanation of Sundberg and Bartholo-
10) Frequently in the spectra of the Western mew, based upon studies of the Western operatic
operatic voice five or six formants were present, but voice, that larynx lowering is responsible for the
in Chinese and early music singing voice there were singing formant and bright timbre, does not acount
four or five formants. In both cases the high for- for their presence in this study. Therefore, lowering
mants, except F5, generally were compacted (see the larynx is not necessary to produce bright timbre
Figs. 2 and 3). Also, the extra peak did not always and its associated acoustic features. It seems that
appear. generally the lower the larynx and the wider the
pharynx, the lower most formants are likely to be
5. DISCUSSION and the easier it is to associate a voice with dark
Here we restate some points of this author's previ- timbre. Moreover, in voice training we know also
ous presentations,"-") to extend the discussion on that one way voice students begin to learn covering
the singer's formant. is often by lowering the larynx and expanding (direct-
1) Although both high and low laryngeal styles ly or indirectly) the pharynx. Usually the voice is
of singing are capable of producing bright voice, dark without much brightness. This step is not very
there are differences in timbre perceptible to trained difficult for them to achieve. It is only in the ad-
ears as we know from voice experience. The differ- vanced stage of voice training, i.e., when the singer
ence of timbre between high laryngeal singing (open) can open the pharynx properly and coordinate this
and low laryngeal singing (covered) relates also to with articulation of the whole vocal tract including
phonetic (vowel) modulation caused by articulation. the front cavity and also with the breathing, that the
For example, the /A/ sung by covered technique in singer can achieve the covered bright voice. This step
high pitch is not a pure /A/ but a "hybrid" vowel of however is a very hard one for them to achieve.
/A/and/o/,(or/A/and/u/). Therefore the formants, Perhaps, the physiological adjustment in covered
especially, the first and second formants (F1 and F2) singing technique that causes bright timbre and its
are correspondingly affected by phonetic (vowel) associated acoustic features is something much more
modulation. In this paper we mostly discuss the complex than the lowering of the larynx. Further
singer's high formants, and we will present and dis- and in-depth research is required to discover the
cuss the results of singer's low formants in another actual physiological mechanics in the production of
paper. bright covered voice.
2) This study does not confirm previous findings Now that trained singers with either high or low
that singing with a high larynx is a characteristic that laryngeal singing technique can produce quality voice
distinguishes an untrained singer from a trained one. and have its corresponding featured formant, the
This result indicates that singing with a high larynx, lowering larynx therefore would not be able to reveal
in and of itself, does not necessarily produce a voice the essence of the production for the professional
of poor quality, or cause poor vocal health. Although singing voice. In fact many singers just keep their
we know from voice training that there are many comfortable and relatively stable laryngeal position
untrained, or poorly trained singers who raise their (of neutral, low or high) to develop quality voice in
larynges in singing, the mechanism of singing with the various styles. This is popular and effective in
a high larynx used by some well-trained singers must voice training as well.

308
 S. WANG: SINGER'S HIGH FORMANTS

However, from the results of our experiments, a voice the raising of the larynx causes the vocal tract
lowered larynx position can be yet regarded as a to shorten which in turn causes some formants of the
distinctive feature for the covered singing voice, singing voice to shift relatively upward if the larynx
though not for the trained voice. Therefore, this position of the speaking voice is lower than that of
physiological feature could be a clue that leads us to the singing voice. In this case the high formant F5
explore the singing technique associated with the may shift over the cutoff frequency so that we only
production of covered timbre. have the first four normal formants plus an extra
3) The study showed that in different styles of formant fe, thus five formants in total-that is, F1,
singing the high formants pattern varied but all had F2, F3, fe, F4, in which fe is between or near F3 and
similar characteristics. The two to four very high and F4 (see Fig. 2). Therefore F3,fe, F4, are compact and
often compact energy envelope peaks including a enhance each other in amplitude so that this five
possible extra peak roughly between 1.8 and 3.8 kHz formants' pattern differs from that in the normal
(BFR) would be the correspondingly similar acoustic speech spectrum. Sometimes in high laryngeal sing-
features in the spectrum, and thus could be called ing, for example, when the larynx is only raised a
the singer's (high) formant. We may believe that if small bit (yet higher than the larynx position in the
there is an extra formant fe between or near F3 and corresponding speaking phonation), and owing to
F4 it could enhance the compactness and thus the some compensation of articulation, F5 is shifted up
amplitude of the peaks F3 and F4, and increase the but is still under the cutoff frequency so we could
sense of brightness. However the bright singing voice also have six formants, that is, F1, F2, F3, fe, F4, F5
can also correspond to three or even two of those (see Fig. 9), which for the same above-mentioned
formants without the extra formant fe (see Fig. 4). reasons are compacted and enhanced in amplitude.
The frequency range (1.8 - 3.8 kHz) of the variations Usually, F5 is not very close to the compact F3, fe,
on the common pattern of the singer's formant falls F4, has a low amplitude, and thus does not appear as
within the most sensitive frequency range (1 - 4 kHz) consistently as F3, F4, as in the similar case with the
of the human ear. This characteristic of singer's speech spectrum. So, generally we found five for-
high formant combined with that of low formants mants instead of six formants, but these five formants
for a singing voice might cause the voice to sound are different from those in the speech spectrum.
brighter and louder, and so please the human ear. Here in the high laryngeal singing voice the five
4) In spectra analysis, we found that in the low formants are F1, F2, F3,fe, F4 (see Fig. 2), but in the
laryngeal singing (covered) voice more often there speaking voice the five formants are F1, F2, F3, F4,
were 5 or 6 formants; however, in the high laryngeal F5. Owing to the difference of the compactness and
(Chinese and early music) singing voice more often the amplitude of the high formant peaks between the
there were 4 or 5 formants. Here we need to explain two cases, the compact F3, fe, F4, in the singing spec-
the number of the formants and the compactness of trum is obviously distinguishable from F3, F4, F5 in
the high formants in the spectra of the different sing-
ing voices. In fact we have to sublate the theory of
lowering the larynx and then the extra formant fe
could be used for interpreting the different patterns
of singer's formant in both high and low laryngeal
singing.
In the covered voice lowering the larynx causes the
lengthening of the vocal tract which in turn causes
most formants of singing generally to shift downward
from the corresponding formants of the speaking
voice. Therefore under the cutoff frequency (5 kHz),
there could be five normal shifted formants plus the
extra formant fe, thus six formants in total-that is,
F1, F2, F3,fe, F4, F5, in which F3,fe, F4, are compacted
owing to the insertion of fe between or near F3 and F4 Fig. 9 Chinese singing voice. sung vowel:
(see Fig. 3). However in the high laryngeal singing /i/,pitch:E4(330  Hz),  LPC  spectrum.

309
 J. Acoust. Soc. Jpn. (E) 7, 6 (1986)

Fig. 11 Sung vowel IA/, a Chinese singing

Fig. 10 Speaking phonation /A/, fundamen- voice, pitch: C4 (261 Hz), LPC spectrum.
tal frequency 131 Hz, LPC spectrum.

the speaking spectrum (see Figs. 6 and 10) though

both of them have the same number of formants.
Moreover, sometimes there are only two formants
in the BFR. It may mean that no extra formant is
produced. That is, on the spectrum of singing voice
the formant could be F1, F2, F3, F4 or F1, F2, F3, F4,
F5. In these cases, the high formants are not very
compact but have a higher amplitude (see Figs. 4 and
11) that make the singer's high formants differ from
speech high formants. However, for some singers,
if the manner of speaking is similar to that of the
singing voice, the spectra of the speaking and singing
could also be similar. For example, among our Fig. 12 A Western opera singer. speaking
subjects, one's speech was very much like his singing phonation /AI, fundamental frequency
("focused"and bright), and the singer's formant 128 Hz, LPC spectrum.
(even including the fe) was found in his speaking
spectrum (see Figs. 6 and 12). In this sense, there is
not an absolute boundary between them. So the Perhaps, this is a good reason why /i/ (and /A/) gen-
formant patterns, or the acoustic features in radiated erally sound brighter than /u/ in singing (and in
spectra between singing and speaking, are generally speaking also). On the other side since /u/ has more
different, but similar and uniform under special stronger formants in the lower frequency range
conditions. Perhaps this is a kind of acoustic ex- (under 1 kHz) than /i/ or /A/ then it sounds darker
planation for the saying "singing is the extension of than them.
speaking". From the results and discussion above we know
We found that in the BFR for /i/ spectrum there that the extra formant could also appear in high
were more peaks than for the fu/ spectrum. It seems laryngeal singing. We may assume that the area
that on the spectra of the /i/ singing phonation, the mismatch and the separate larynx resonator (as
extra formant occured more often than that of /u/. stated in Sundberg's explanation) could happen in
We are also aware that generally the larynx position high laryngeal singing. This result and assumption
when singing /i/ was higher than that when singing made the maximum area ratio 1/6 questionable. It
/u/!(See  Fig.1).  Moreover  for/i/,  the  frequency  of
seems necessary for future discussion to have a
F2 was very high and close (even it joins) to the com- further investigation of the actual ratio value of the
pacted (and separate) F3, fe, F4 in BFR, and they area in both high and low laryngeal singing, and of
enhanced each other, causing a high amplitude. the maximum ratio within which condition the larynx

310
 S. WANG: SINGER'S HIGH FORMANTS

tube separates from the remainder of the vocal tract. frequency which is determined by the separated
So far we have explained several cases of singer's larynx tube resonator actually has three different defi-
high formant in high and low laryngeal singing. In nitions and values : (1) F4; (2) the extra formant fa
fact the singer's formant patterns can be more varied which, as mentioned in the introduction, is between
and complex than the above-mentioned cases, and but separate from F3 and F4; (3) a cluster of F3, F4,
affected by many factors. Because the extra formant F5. The concept of the singing formant is confused
determined by the independent larynx tube is not not only by changing meanings but also by the princi-
always located between F3 and F4, and F5 does not ples of acoustics. In fact Sundberg's recent definition
always show up, the pattern of the singer's high of singing formant, that is the cluster of F3, F4, F5
formant in the spectrum sometimes is very intricate. appears not only inconsistent but also incompatible
The rounding or spreading and the protuding or with his explanation. From the acoustic viewpoint
withdrawing of the lips effect the length of the vocal of voice production F3, F4, and F5 cannot all be pro-
tract quite a bit-this is another reason that some- ducted by this separate larynx tube. Moreover, this
times we can not entirely explain the complex shift- definition would mean that F3, F4 and F5 would be
ing of the formant only by the change of the larynx regarded as one formant, and therefore, there would
location. In low larynx singing if the length of the be only three formants in total in the singing spectra
rest of the vocal tract except the independent lar-
(i.e., F1, F2 and the cluster). Obviously this is not an
ynx tube is shorter than that of the corresponding inclusive picture of the various patterns of the
speaking vocal tract the shifting of the formants in singer's formant as we have shown in the figures,
the singing voice will not be as simple as what Sund- especially, as F5 usually is not close to the cluster
berg described. All the above factors have to be (F5 - F3 can be as big as 2 to 2.5 kHz).
considered when analyzing spectra. Moreover the Today Sundberg's terms (singing formant and
modification of the vowel, the compensation and singer's formant), and the meanings (fe and the cluster
predominance of articulation, the morphology of the of F3, F4, F5) seem confused or appear vague as used
individual vocal tract, etc., greatly effect the formants, in the literature. For example, the meaning used
especially the two low formants Fl and F2. in singer's formant in SMAC 83 Proceedings and in
5) It is noticed that Sundberg4 5,13, 20 ,21) changed The Science of Sound") seem different. It is neces-
the meaning of the singing formant. At first, Sund- sary to clarify the concept of Sundberg's singer's
berg13) wrote "The singing formant, located in the formant for better understanding and future discus-
frequency range 2.5•`3.0 kHz is strongly dependent sion. Since the extra peak fa (other than F3 and F4)
on the larynx tube. Actually, the larynx tube appears does not always appear in the spectra of singing voice
to behave as a Helmholtz resonator for the fourth it would not be proper to consider it a criterion of
formant." According to his explanation,4 5) the voice quality (evaluation) and to use it as the singer's
singing formant, as the frequency of the independent formant to generalize the acoustic feature in the
larynx tube, should equal F4. Recently Sundberg20,21) radiated spectrum of singing voice. On the other
has asserted that the "singing formant or the singer's hand, because sometimes the extra peak fa appears,
formant, as I prefer to call it today, ... is constituted and F5 does not always appear (such as in the above-
by a clustering of F3, F4, and F5, ... and this cluster mentioned high laryngeal singing, the F5 could be
can be explained as an extra formant associated shifted over the cutoff frequency, and even when F5
with the larynx." [Sundberg13) also said that the appears it is not usually close to the cluster F3, fa,
singing formant "seems to consist of two or three F4), it does not seem proper to use a cluster of F3,
formants." This definition of his would not be com- F4, F5, as the definition of the singer's formant either.
patible with his explanation of F413) and fe4 , 5) as- Otherwise, not only would they ignore the variety of
sociated with a separate larynx tube. Now that formant patterns of the singing voice which charac-
he has changed it to "a clustering of F3, F4, F5," terize individual voice timbre but would put forth
we only discuss here his new one.] Here logically the questionable conclusion that the soprano voice
we understand Sundberg's definition of singing has "no singing formant" (although to date we have
formant as follows : First, in Sundberg's terms not developed an accurate means of measuring the
singing formant and singer's formant are the same. formants of the female voice in high fundamental
Second, his concept of the singing formant and its frequency). During singing a soprano does adjust the

311
 J. Acoust. Soc. Jpn. (E) 7, 6 (1986)

shape and length of her vocal tract, and therefore the formant frequencies of a raised larynx are,•@ ... a com-

corresponding acoustic features (natural frequencies) bined rise in several vowels of the third and fourth

of the modulated vocal tract must exist (some•@ re- formant frequencies." However, in this research for

searchers"' also reported some findings of singer's the professional high laryngeal singing voice, general-

formant in soprano singing voice though their con- ly the formants except F4, tend to be shifted upward

cept about singing formant is a bit different). This relative to the corresponding formants in the speak-

author also analyzed some tenors' falsetto voice in ing voice. The shifting direction of formants in high

whose spectra there does not seem to be very high laryngeal singing are not as consistent as those in

energy peaks in the•@ BFR. It is•@known3,22,23) that the low laryngeal singing. One explanation for this could

male full voice is produced mainly with the heavy be that the vowel modification in high laryngeal sing-

mechanism and the male falsetto voice with the light ing is not as great as that in covered (i.e., low laryn-

mechanism. The soprano (especially in high pitch) geal) singing. Another explanation for this could be

uses light mechanism (mixing some heavy mechanism that the larynx position when speaking is not usually

in singing). Apparently it is the vibrating mode of the physiological resting position, and is mostly

the vocal cords (asociated with certain adjustments higher than the physiological resting position. There-

of supra/sub glottal systems) rather than the voice fore the laryngeal location difference between high

type or the sex of the singer that effect the amplitude laryngeal singing and speaking is not as large as the

of the high harmonics and thus of the high formants, difference between lowered laryngeal singing and

and so effect the singer's formant. Moreover, since speaking. Sometimes the larynx location of speaking

the amplitudes of all the formants affect the dynamic was found to be even higher than that of high laryn-

level of the singing voice the singer's (high and low)

geal singing for some subjects, especially in the mid-
formants also do. dle range. So the direction (upward/downward) of

Sundberg's concept and explanation of singing the shifted formants in singing vs. speaking roughly

formant due to the lowered larynx was limited to depends on the relative location of the larynx in

male voice, and•@ he4,5) has a separate and different singing vs. speaking. That is, the shifting direc-

conclusion and explanation for the female voice. tion of the formants could roughly coincide with the

Now in some soprano voices Sundberg has found direction of the larynx from speaking to singing. We

raised larynges during singing. This could also raise should also notice that F4 in high laryngeal singing

a logical inference•@ question•@ : Would this new finding is not likely shifted up from that in speaking. Clearly
in sopranos be compatible or irrelative to Sundberg's the F4 behavior in high laryngeal singing could not

lowered larynx explanation for the singing formant

generally be explained by the larynx position, i.e., the
in the male•@ voice,20) to the conclusion that•@ soprano shortening of the length of the vocal tract. So com-

has no singing formant, and to his former•@ conclu-

pensation of articulation and other reasons should
sion"' that "singing with a high larynx may damage be considered in order to explain the frequency differ-

the vocal cords" ? ence of formants in singing and speaking. Another

Based upon the results of this study and combined example•@ : for the low formants F1•@of•@/i/ in covered (low

with some results of other researchers (including larynx) singing is actually higher (rather than lower)

Sundberg), the definition of the singer's formant than that in the corresponding speaking. Obviously

proposed in point 3 of this paper would not have the this can not be explained by the lengthening of vocal

above problems, but comments and criticism are tract. Perhaps, this is predominantly caused by vowel

expected and welcomed. modification. That is,•@/i/ is modified•@ by•@/u/. We know

6) In line with the explanation of the lowered that the•@ F1 of•@/u/•@ is higher than•@ /i/ in speaking voice.

larynx,•@ Sundberg4,5'24) claimed that the formant in So the F1 of /i/ in covered singing voice was shifted

the singing voice shifted downward all formant fre- up. It seems that the shifting of low formants from

quencies other than the larynx-dependent extra speaking to singing is sensitive to the change both of

formant relative to the corresponding formants in the shape and length of the vocal tract, and that of

the speaking voice. Sundberg25) concluded that "the high formants to the change of the length, i.e. mainly

first order effect on the formant frequencies of a shift by the larynx position.

in the larynx position stems from the•@ shortening/ 7) We know that the function and characteristics

lengthening of the pharynx cavity. The effects on the of voice source and the interactions between the

312
 S. WANG : SINGER'S HIGH FORMANTS

source and the supra/sub glottal system are very various formant patterns of singing voice associated
important in affecting voice quality, especially in high with different laryngeal location and types. The low
voice. In voice training, in fact one of the main tasks laryngeal position still can be regarded as a distinc-
for the singer is to discover the sense of proper tive feature for the covered voice though not for the
interactions between the larynx and breathing (sub- trained voice; the source characteristics, and the
glottal system), and between the larynx and articula- means of handing the relationship between the source
tion (supra glottal system) in order to be capable of and the supra/sub glottal systems are possibly differ-
producing a good voice by correctly handling the ent in styles of singing associated with high/low laryn-
relationship. geal singing.
The different laryngeal locations during phonation
in different styles of singing perhaps means different ACKNOWLEDGEMENTS
functions or characteristics of voice source; in high The author expresses thanks to his former advisers
laryngeal singing, the voice source adjustment used and colleagues at the Speech Communication Group
by well trained singers and untrained singers may also of Research Laboratory of Electronics at the Mas-
be different. For example, since the amplitude of sachusetts Institute of Technology, and his colleagues
high formants of good high laryngeal singing voices at the Physics Department and the Music Depart-
is high, by inference, the harmonics and so the fun- ment, York University. Special thanks are due K.
damental, in their voice source must be also strong Stevens, G. Fant, O. Brown, J. McKay and J. Burt.
in amplitude. Therefore the physiological mechanism Gratitude is expressed to W. Wang, and K. Komar,
of the voice source for them is possibly different from to singers in China, Norway and the USA who were
that for poor high laryngeal singing. The relation- the subjects of this study, and to nineteen teachers
ship between the source and supra/sub glottal sys- and singers who served as jury in the perceived timbre
tems (such as lung pressure, glottal maximum open- study. Many thanks go to R. Kelly, C. Sharpe and
ing, closure speed, degree of discontinuity at closure R. Witmer.
together with articulation, moreover, glottal friction The research and presentation were supported in
and inductance, etc.) might vary with different styles, part by the People's Republic of China and the (US)
types, and even individual singers, and then, the Voice Foundation, York University, and Brown
approach for handling this relationship should also Voice Seminar.
be different. That might be one of the reasons why
we can find good and bad singing in either high and REFERENCES
low laryngeal singing, and so laryngeal height should 1) W. Bartholomew, "A physical definition of good
not be considered a reliable means of distinguishing voice quality in the male voice," J. Acoust. Soc. Am.
good from poor singing voices. Further investigation 6, 25-33 (1934).
2) S. N. Rzhevkin, "Certain results of the analysis of a
of the voice source and the interactions between the
singer's voice," Sov. Phys. Acoust. 2, 215-220 (1956).
laryngeal adjustment with supra/sub glottal systems
3) W. Vennard, Singing : The Mechanism and the Tech-
will shed insight into the physiological mechanism nic (Carl-Fisher, New York, 1967).
and acoustic features of singing voice associated with 4) J. Sundberg, "Articulatory interpretation of the
different larynx position. singing formant," J. Acoust. Soc. Am. 55, 838-841
(1974).
6. CONCLUSION 5) J. Sundberg, "The acoustics of the singing voice,"
Sci. Am. 3, 82-91 (1977).
Professional singers produce singing voice in style
6) W. T. Bartholomew, "The role of imagery in voice
associated with different laryngeal height. The teaching," NATS, USA, Research Committee, No. 1
singer's (high) formant patterns are varied but have (1951).
similar characteristics. The extra formant is also 7) T. Shipp and K. Izdebski, "Vocal frequency and
found in the high laryngeal singing voice, and it is vertical larynx position by singers and nonsingers,"
not a consistent acoustic feature in all professional J. Acoust. Soc. Am. 58, 1104-1106 (1975).
8) W. Frummhold and G. Hoppe, "Tomographische
singing voices. Lowering the larynx does not explain
Studien zur Funktion des menschlichen Kehlkopfes.
the extra formant, singer's formant and voice quality
1. Mittellung : Unterschiede in der Stimmlippen-
(brightness and evaluation). Sundberg's two defini- mechanik," Folia Phoniatr. 17, 83-91 (1965); 18,
tions for singer's formant are not inclusive for the 81-90 (1966).

313
 J. Acoust. Soc. Jpn. (E) 7, 6 (1986)

9) K. Faaborg-Anderson and W. Vennard, "Electro- 18) W. Seidner, H. K. Schutte, J. Wendler, and A.

myography of extrinsic laryngeal muscles during Rauhaut, "Dependence of the high singing formant
phonation of different vowels," Ann. Otolaryngol. on pitch and vowel in different voice types,"
Reinlaryngol. Laryngol. 73, 248 ff (1964). SMAC 83, Proc. of the Stockholm Music Acous-
10) R. Luchsinger and G. E. Arnold, Lehrbuch der tics Conference Publications, Royal Swedish Acade-
Stimm- and Sprach-Herkunde (Springer-Verlag, my of Music, No. 46: 1 (1985).
Vienna, 1959). 19) A. Johnsson, J. Sundberg, and H. Wilbrand, "X-
11) W. Ruth, "The registers of the singing voice," NATS ray study of articulation and formants frequencies
Bull. May 2-5 (1963). in two female singers," SMAC 83, Proc. of the
12) A. Soninen, "Paratasis-gram of the vocal cords and Stockholm Music Acoustics Conference Publica-
the dimensions of the voice," Proc. 4th Int. Congr. tions, Royal Swedish Academy of Music, No. 46:
of Phonetic Science, The Hugue : Mouton (1962). 1 (1983).
13) J. Sundberg, "Formant structure and articulation of 20) J. Sundberg, Letter to S. Wang, Sundberg also
spoken and sung vowels," Folia Phoniatr. 22 (1), noticed Wang that he realized recently "that singing
28-48 (1970). with a high larynx occurs in professional singing
14) S. Wang, "Bright timbre, acoustic features and and does not have to be harmful to the voice func-
larynx position. (video film)," 12th Brown Voice tion" (1984).
Seminar, Amherst, USA (1983); 14th Symposium : 21) J. Sundberg, "The voice as a sound generator,"
Care of the Professional Voice, Denver, USA (1985). Research Aspects on Singing Publications issued by
15) S. Wang, "The relationship between bright timbre, the Royal Swedish Academy of Music, No. 33
acoustic features and larynx position: a comparsion (1981).
of Western operatic with Chinese and some Western 22) T. Rossing, The Science of Sound (Addison-Wesley,
early music singing styles" (1983) (This paper was Reading Mass., 1982).
presented at the 105th Conference of ASA and in- 23) J. C. Wilcox, The Living Voice (Carl-Fisher, New
cluded in NATS Bull. Jan/Feb, USA (1985)). York, 1945).
16) S. Wang, An Introduction to Voice Production (Qu- 24) B. Lindbolm and J. Sundberg, "Acoustical conse-
Fu Teachers College Textbook, Qu Fu, 1976) (in quences of lip, jaw and larynx movement," J. Acoust.
Chinese). Soc. Am. 50, 1166-1178 (1971).
17) G. Troup and H. Luke, "Some radiological observa- 25) J. Sundberg and P. E. Norstrom, "Raised and low-
tion of vocal source-vocal interactions," SMAC83, ered larynx-The effect of vowel formant frequen-
Proc. of the Stockholm Music Acoustics Conference cies," Speech Transmission Lab. Q. Prog. Status
Publications, Royal Swedish Academy of Music, Rep. 2-3, 35-39 (1976).
No. 46: 1 (1983).

314