It’s Not Over Until the Fat Lady Speaks: Formant Variation in Operatic Singing and Speech

 

Some, but still very little, study has been done in the way of examining the differences

between singing and speech. What can be distinguished very clearly by the ear becomes more

problematic when examined by spectrogram. When examining trained classical singers, a few

clear differences arise, such as the presence of the so-called “singer’s formant” and vibrato.

However, when discussing technique with trained operatic singers, the way they form their

words—particularly vowels—with their vocal tracts differs radically from everyday speech. This

raises the question: when operatic singers apply similar techniques to speech to modify their

speaking voices—hereby known as “operatic speech”—are their mannerisms more speech-like

or song-like? By recording three operatically trained tenors in several seconds of singing, speech,

and operatic speech, I examined the differences and similarities between all three forms,

specifically focusing on formants in the vowels. I hypothesized that when trained singers spoke

operatically, singer’s formant would be present and the formants on their vowels would more

closely mirror singing than normal speech.

Research has led to the discovery of a phenomenon called “singer’s formant.” Present

only in male singing voices, singer’s formant is a resonance that occurs around 3000 Hz. When

an opera singer performs, they are often singing accompanied by a full orchestra. A single man’s

voice cannot compete with dozens of instruments, so they must change the way they produce

sound in order to be heard over the orchestra without straining their vocal folds. The weakest

part of the orchestral spectrum is about 3000 Hz, so a singer can change the formation of his

vocal tract to cluster the third, fourth, and fifth formants to produce the resonance in the

orchestra’s weak spot and be better heard. In order to make this change, a singer can lower his

larynx, widen his pharynx, laryngeal ventricle, and his piriform sinuses located above the vocal

folds. (Kreiman 2011)

Singer’s formant occurs in all male voice parts, but at varying frequencies. For example,

most professional basses have a singer’s formant that falls in a range of about 2300 Hz – 2500

Hz, most baritones’ are between 2500 Hz and 2700 Hz, and tenors tend to fall between 2700 Hz

– 2900 Hz. The differences between singer’s formant in voice part can likely be attributed to

vocal tract length. Basses tend to have the longest vocal tracts, between 23.5 and 25 cm,

baritones tend to have vocal tracts between 21.5 and 24 cm, and tenors often fall between 19 cm

and 22 cm (Dmitriev 1979). The longer the tract, the lower the voice, and therefore, the lowering

of the larynx has different effects on the frequency of the singer’s formant. As for women, altos

have a much weaker singer’s formant than men, and sopranos have no singer’s formant at all.

Interestingly enough, famous sopranos have more energy in these high frequencies than less

famous sopranos, implying that a high-pitched female voice that carries these frequencies may be

the key to success. In addition to all the voice parts, it was found that untrained non-singers also

do not have a singer’s formant, meaning that this frequency is found through training (Sundberg

1990).

This mysterious frequency has been the subject of much question and study. As scientists

and linguists dig into the subject, some questions have been answered and others remain. For

example, one study took ten acclaimed tenors trained in either a Western or Chinese style, had

them sing the vowels /ɑ/, /i/, and /u/, and had 19 voice teachers rate each vowel on a one to seven

scale of brightness. A narrow strip of colored tape was placed on each singer’s neck at a right

angle to the airway and the singer was filmed by the researchers. A computer analyzed the vocal

recordings for acoustic features. The results found that all subjects’ voices were reported to be in

the “bright range” with varying degrees. Across the board, /i/ and /ɑ/ were judged to be brighter

than /u/. For the Chinese-style tenors, the larynx generally was raised above its normal position

while singing, especially when they sang in their higher range, and the Western singers’ larynges

were usually below its normal resting position. This rule was even true for the singers who could

sing in both styles: when they sang Chinese, the larynx was raised, and when they sang Western,

the larynx was lowered. Furthermore, when the pitch increased in the Chinese singers, the larynx

moved up, and the opposite occurred in the Western singers. All of the formants’ frequencies

were higher in the Chinese style than in the Western style. The Western singers’ sung vowel

phonations were often lower than their spoken, and there were five or six formants present in the

Western vowels and only four or five in the Chinese style. In both cases, the upper formants

demonstrated clustering. (Wang 1986) This study was relevant to my experiment because it

explains some of the phenomena on the three western tenors’ spectrograms and measurements.

In most cases for most formants, their speech phonations were higher than their sung ones, and

any that deviated from this rule could be explained by a small sample size, natural variation, and

potentially any incorrect readings from the software (though I accounted for obvious machine

mistakes by finding the approximate frequency manually). This also notes five or six formants in

each vowel for each singer, which I also noted, though I only recorded the first four formants for

each vowel.

What else differentiates singing from speech? An important difference to note is

amplitude. When speaking casually, one only projects his voice to be heard by those around him,

but a singer must often sing loud enough to fill an entire auditorium or theatre and be heard over

a full orchestra. Singer’s formant and differently shaped vowels help produce amplitude, but

several other factors play into it as well. Higher sub-glottal pressure reflects louder amplitude,

though higher pitch also requires higher sub-glottal pressure. When speech was evaluated for

subglottal pressure as well, it was found that nonsense speech produced practically no sub-glottal

pressure, but emotional stresses on certain important syllables were created using this pressure as

well. Normally, loudness in speech is controlled by shifts in fundamental frequency, so

amplitude produced by sub-glottal pressure in speech is better characterized by “stress.” Pitch

passively rises with amplitude when speaking due to the shift in fundamental frequency, which

cannot happen if a singer wishes to sing different pitches at a loud amplitude. Therefore, it makes

sense that they would need to use sub-glottal pressure to increase amplitude. (Sundberg 1990)

Another notable difference between speech and operatic singing is the way that singers

form vowels. In everyday speech, the mouth’s goal is to find the most efficient way to form each

sound. However, singing often has extended vowels and efficiency is less important than

amplitude. For example, the most efficient way to say the “corner vowel,” or a vowel created in

an extreme corner of the mouth (front high, front low, back high, back low), /i/ is to lift the

tongue up to the front upper edge of the mouth. The resonant chamber in the oral cavity then is

located behind the tongue, so it carries less sound than another corner vowel, /ɑ/, which is

located at the very back and bottom of the mouth. When making the vowel /ɑ/, the resonant

chamber in the oral cavity is located primarily in front of the tongue and allows for more sound

to be heard. In order to make every vowel clear, operatic singers are taught to make vowels such

as /i/ in a different way. To make /i/, a singer will lift the middle-back of the tongue to the roof of

the mouth. This sounds the same as a front high /i/ to a listener, but is more amplified and can

better be heard over an orchestra in a large auditorium. (McCreary)

Due to the specific example of how to differently form vowels, I decided to test the

differences between vowels when sung operatically, spoken normally, and spoken with an

operatic tone to the voice. The two areas on which I focused were the presence of “singer’s

formant” in sung and operatically spoken text and also the differences in formant clustering

between the all three ways of delivering the lyrics.

 

To perform this experiment, I selected three tenors to record. The first, Connor

McCreary, is a 23-year old tenor who recently graduated the Thornton School of Music at the

University of Southern California’s undergraduate program in vocal performance. The second,

Anthony Moreno, is a 30-year old tenor currently enrolled in Thornton’s doctoral program, and

the third, Kyle Chase, a 25-year old tenor enrolled in the same doctoral program. Each of them

choose an English-language classical song in their repertoire so they would be familiar with the

lyrics and be able to sing it to the best of their abilities. Each subject then 1) sang the line of their

song, 2) spoke the same lyrics in a normal speaking voice, and 3) spoke the lyrics in an operatic

singing voice. They then recorded themselves speaking freely, describing what they were doing

with their vocal tracts when singing and speaking operatically. I then examined the spectrograms

of each subject, especially taking note of their vowels and formants.

First, I selected either the entire line they chose or part of the line and counted how many

vowels were present. Kyle’s lyric of choice, “When the air sings of summer I must wander

again” had 13 vowels present when counting diphthongs as one vowel. Connor’s lyric, “Come

under the shadow of this gray rock” had 10 vowels when considering diphthongs one vowel, and

Anthony’s lyric, “In the evening as far as the eye can see, herds of black pianos” had 17 under

the same condition. Closely examining each vowel, it became clear that all of Connor’s vowels

had a formant present at 3000Hz, all but one of Anthony’s did, and all but two of Kyle’s did as

well. When looking at the normally spoken versions of these three lyrics, Anthony and Kyle had

no formants that could be mistaken for a singer’s formant, and while Connor’s reading had many

formants at 3000Hz, his free speech showed several formants at 3000Hz on various vowels,

meaning his natural speaking voice happened to have some formants in this range. Finally, when

examining each operatically spoken version of the lyric, it became clear that for every vowel that

had a singer’s formant in the sung version, the same formant was present in the operatically

spoken version.

 

Figure 1: Anthony’s singer’s formant for the lyric “In the evening as far as the eye can see, herds of black pianos”

Figure 2: Connor’s singer’s formant for the lyric, “Come under the shadow of this gray rock”

Figure 3: Kyle’s singer’s formant for the lyric “When the air sings of summer I must wander again”

Figure 4: Anthony’s singer formant present when operatically speaking the lyric, “In the evening as far as the eye can see, herds

of black pianos”

Figure 5: Connor’s singer’s formant present when operatically speaking the lyric, “Come under the shadow of this gray rock”

Figure 6: Kyle’s singer’s formant present when operatically speaking the lyric, “When the air sings of summer I must wander

again”

Another difference of interest between operatic speech and singing is generally the

placement and clustering of formants. To test this, I picked all the corner vowels present in the

three chosen lyrics: /i/, /æ/, and /ɑ/. In none of the three songs did the corner vowel /u/ appear.

After finding the first, second, third, and fourth formant for every clear instance of each vowel, I

averaged the data together to find a mean F1, F2, F3, and F4 for every vowel when sung, spoken,

and spoken operatically. Finding a mean helps to account for the variability that occurs naturally

between speakers and contexts of the vowel. I then compared each vowel across singing, speech,

and operatic speech, using the average spoken formants from a 1995 study (Hillenbrand 1995) as

a control.

Figure 7: All recorded numbers are in Hz and “average” refers to Hillenbrand 1995

When the formants for each vowel were compared to one another across speech, singing,

and operatic speech, there was no pattern for formant similarity across all vowel types. However,

for all four formants for /i/, speech and operatic speech had closer frequencies. For /æ/, speech

and operatic speech had more similar frequencies for F2 and F3, and for /ɑ/, speech and singing

had closer F1 and F2s and singing and operatic speech had more similar frequencies for F3 and

F4. Therefore, it can be concluded that singing and operatic voice do not share more similar

formant frequencies than operatic voice and speech or speech and singing.

One important aspect of singer’s formant is the clustering of F3, F4, and F5. To examine

whether or not this clustering occurs during operatic speech, I subtracted F3 from F4 for all three

delivery types and compared the differences as follows:

 

Formant 4 Minus Formant 3

                                    /i/               /æ/              /ɑ/

Speech                       824.40      897.41         800.39

Singing                      579.84      565.815        531.93

Operatic Speech      668.33      601.75          477.24

 

As can be seen here, there is a difference of about 800-900 Hz for all corner vowels when

these three men speak them. However, there is only a difference of 477-668 Hz when singing or

speaking them operatically. These differences provide evidence that F3 and F4 are clustered

closer together when singing or speaking operatically, which also helps to prove that the

resonance seen in the spectrograms for operatic speech is, in fact, singer’s formant.

The fact singer’s formant and formant clusters are present in operatic speech reveals

several facts about what operatic speech truly is. In order to speak operatically, the subject

lowered his larynx and widened his pharynx just as he would to sing over an orchestra. Another

revelation about what a singer must do to speak operatically relies on amplitude. As evidenced

by this waveform of Anthony’s singing, operatic speech, and normal speech, respectively,

singing has the highest amplitude, normal speech has the lowest, and operatic speech is

somewhere in the middle.

Figure 8: Anthony’s amplitude

Is this change in amplitude due to sub-glottal pressure or change in fundamental

frequency? Upon reviewing the spectral slice for the syllable /o/ of “pianos,” I found an

approximate F0 of 300 Hz for the sung syllable, an F0 of about 100 Hz for the operatically

spoken syllable, and about 90 Hz for the spoken syllable. As the fundamental frequencies are

quite similar between the two spoken syllables, it can be assumed that the reason for operatic

speech being louder than normal speech is due to the same sub-glottal pressures that, in addition

to higher fundamental frequencies, help produce high amplitudes in operatic singing. Therefore,

sub-glottal pressure can be added to the list of changes the subjects made to their vocal tracts to

produce operatic speech.

If all of these changes to the vocal tract can be made when speaking, what then

differentiates operatic speech from operatic singing? One difference is length of time. Many

songs require certain notes be sustained for a length of time, and generally, each syllable is given

more time in song than in speech. For example, when Connor sings the syllable /ʃæd/ of the word

“shadow,” the syllable is 1.44 seconds long. When he speaks the same syllable, it is only .35

seconds in length. When researchers sought to convert recorded speech into synthesized song,

duration of phonemes was one of the manipulated variables (Saitou 2007). The other acoustic

features they manipulated that represent the difference between speech and song are the

fundamental frequency and spectrum. The fundamental frequency represents pitch, which varies

in natural speech but varies even more in music as each pitch represents a musical note. They

also identified four other factors present in the fundamental frequency that differentiated between

speech and song: “1. Overshoot: a deflection exceeding the target note after a note change. 2.

Vibrato: a quasi-periodic frequency modulation (4-7 Hz). 3. Preparation: a deflection in the

direction opposite to a note change observed just before the note change. 4. Fine fluctuation: an

irregular frequency fluctuation higher than 10 Hz” (Saitou 2007). Vibrato can clearly be seen in

the singing of each tenor, as represented by the sine-wave like design in the following

spectrogram of Connor’s vowel /æ/ in the syllable /ʃæd/:

Figure 9: Evidence of vibrato in the syllable /ʃæd/

The same sine-wave like pattern is not present in either the normally spoken lyric or the

operatically spoken lyric, meaning that vibrato is a feature unique to singing alone and

differentiates it from operatic speech.

Operatic speech, while not a commonly known or used phenomenon, is an interesting

middle ground between singing and speech. Although speaking operatically does not appear to

change the fundamental frequency of someone’s voice, the singer’s formant and formant

clustering could confuse a linguist trying to match spectrograms of someone’s voice and could

serve as a convincing vocal disguise. The one downfall to this form of disguise is that the only

community that could use it would be operatically trained males, which, unless someone were

trying to solve a bomb threat at an opera house, would likely end up pointing to a suspect. All

this form of speaking requires is a lowered larynx, widened pharynx, and similar sub-glottal

pressure to emotionally stressing a syllable. Someone could potentially learn how to mimic this

technique, though it would take explicit instruction as untrained singers did not exhibit the same

formant clusters and singers’ formants. Overall, this way of manipulating one’s voice proves to

share many similarities with singing, with only vibrato, duration, some amplitude, and some

variations related to fundamental frequency keeping it as speech instead of song. Operatic speech

serves as a sort of bridge between singing and speech, sharing attributes with both and

connecting them together.

 

Laura Russell is a current junior at the University of Southern California pursuing a double-major in Linguistics and Communication. Outside of the classroom, she works as a tutor in the Student-Athlete Academic Services department and participates on the board of Undergraduate Students in Linguistics. After graduation, she plans to attend law school in order to pursue a career in trademark law.

Bibliography

Dmitriev, L., & Kiselev, A. (1979). Relationship between the Formant Structure of Different

Types of Singing Voices and the Dimensions of Supraglottic Cavities. Folia Phoniatrica

Et Logopaedica, 31(4), 238-241. doi:10.1159/000264170

Hillenbrand, J., Getty, L. A., Clark, M. J., & Wheeler, K. (1995, January 17). Acoustic

characteristics of American English vowels. The Journal of the Acoustical Society of

America, 97(5), 3099-3111. doi:10.1121/1.411872

Kreiman, J., & Sidtis, D. (2011). Foundations of voice studies: An interdisciplinary approach to

voice production and perception. Chichester, West Sussex: Wiley-Blackwell.

McCreary, C. (n.d.). Physical Characteristics of Operatically Sung Vowels [Personal interview].

Saitou, T., Goto, M., Unoki, M., & Akagi, M. (2007, October 21-24). Speech-to-Singing

Synthesis: Converting Speaking Voices to Singing Voices by Controlling Acoustic

Features Unique to Singing Voices. 2007 IEEE Workshop on Applications of Signal

Processing to Audio and Acoustics, 215-218. doi:10.1109/aspaa.2007.4393001

Sundberg, J. (1990). What’s so special about singers? Journal of Voice, 4(2), 107-119.

doi:10.1016/s0892-1997(05)80135-3

Wang, S. (1986). Singer’s high formant associated with different larynx position in styles of

singing. Journal of the Acoustical Society of Japan, 7(6), 303-314. doi:10.1250/ast.7.303

---