NOTE: all the voice sound samples were performed by Mariachiara Gallo (mezzo-soprano) on 04/12/2007. Sound engineer: Valerio Carboni.

The voice is a very special "instrument", since all its parts are contained inside our body. Exactly as in the case of the other instruments, it needs an "exciting" element (that is, the air, which comes out thanks to the push of the diaphragm), a vibrating body (in this case the vocal folds, whose tension, length and thickness determine the pitch of the sound) and a "resonant environment" (represented by the rib cage as well as the oral cavity, the nasal cavity and several other small areas distributed in the skull). The incredible peculiarity of the human voice lies in the fact that the main "resonator", the mouth, is movable, and its shape can therefore be changed while the sound is being produced. As a consequence its timbre can be varied continuously (for example, keeping a constant pitch, passing from an "a" to a "u"), just as if we were seamlessly transforming a clarinet into a flute. In the following sections we will try to outline the mechanism by which the voice is produced and modulated, to understand exactly how this timbral morphing takes place.

How the voice is produced

From the lungs to the vocal folds: flow and pressure

1. The force that produces the voice and singing is the force of the muscles of the rib cage. By contracting, they compress the lungs, and so generate a continuous flow of air that rises from the lungs along the trachea. Up to this point no sound wave has yet been generated, and the air flow is similar to the expiratory flow.
2. Towards the upper end of the trachea the air breaks against a sort of "curtain" (the so-called vocal cords, although their shape is more that of vocal folds), whose shape and opening can be adjusted through a complex system of muscles. At rest the vocal folds are in the "open" position (see figure), allowing air to pass for breathing. When they are tensed by the muscles they close to a variable extent, narrowing the passage.
3. The continuous flow of air striking the vocal folds undergoes a sudden increase in pressure but, when the pressure is sufficient to overcome the tension of the folds, the air suddenly passes through the slit, with a corresponding sudden decrease in pressure by the principle of conservation of energy (which in this case is called Bernoulli's theorem).
4. The previous "frame" captures the period of a single oscillation which, from this point on, sets in in the vocal folds and in the air column involving them. The mechanism is therefore entirely analogous to what happens in wind instruments through the embouchure, the player's lips, or the reed.
5. The frequency of the sound produced therefore depends, ultimately, on the oscillation frequency of the vocal folds, which in turn depends on their tension, their density and their length. In adult males the vocal folds are about 17-25 mm long, while in females about 12.5-17.5 mm, which explains the difference in range between males and females, around 125 Hz and 210 Hz respectively.

From the vocal folds upwards: effects of resonance

After the air column has been set vibrating in the larynx, the sound wave still has to pass through several regions inside the human body before coming out into the open. The wave produced by the vibration of the vocal folds is not in itself a pure sound, but contains several harmonics, to an extent that depends on the details of the oscillation. It is the intensity of these component harmonics that is greatly modulated as the wave passes through the remaining parts of the vocal apparatus.

1. Indeed, from the vocal folds upwards, the sound wave essentially passes through some cavities of rather complex shape, and of variable geometry thanks to the changing shape of the soft parts (glottis, tongue, palate).
2. Each cavity is subject to the phenomenon of resonance (see also Helmholtz resonators). The net effect on the incoming sound wave is therefore to damp its spectral components that are off resonance, and to enhance those close to resonance.
3. To modify the timbre of the voice they produce, singers and actors learn many refined techniques. For example, they are able to exploit, besides the resonances of the mouth, those of the nasal fossae and of the paranasal sinuses.

For the same intensity, fundamental frequency and initial timbre, therefore, the voice is still subject to a deep and controllable modification of its spectral content, which makes it possible to articulate a great many sounds in the form of vowels and consonants. The complexity and refinement of the control of all the muscles used in phonation is even more evident when watching the X-ray films presented at the beginning of the page.

Some remarks

1. As long as the vocal folds are kept at constant tension, a note of constant pitch is produced, whose frequency corresponds to the oscillation frequency of the vocal folds.
2. As long as the air flow is kept at constant pressure in the lungs, a note of constant intensity is produced.
3. The amount of air needed to produce the voice is very small compared with that needed for breathing. This is easy to see by breathing out deeply and observing that, even with the lungs "almost empty", one can still produce and modulate sound for several seconds. This observation reflects the fact that the sound wave does not consist in the transport of the fluid (in this case air) that propagates it, but in the oscillation of the medium. In principle, therefore, it would not even be necessary for the air to physically leave the mouth; this happens, however, because at each periodic opening of the vocal folds a little air escapes (a parameter called glottal flow). A well-trained singer is nevertheless able to produce notes even of great intensity without a candle flame placed in front of the mouth moving appreciably. A singer's ability to sustain a sound for a very long time therefore depends more on the training of the breathing reflexes than on lung capacity.
4. The air pressure needed for phonation is also small compared, for example, with that needed to play a wind instrument. A higher pressure, however, to keep the frequency unchanged, must also correspond to a greater tension of the vocal folds. There is therefore an optimal pressure range, determined by the acoustic impedance of the vocal tract.

The mechanism illustrated so far explains how the sound wave is formed, but not yet the incredible variety of sounds that can be produced. These are largely determined by the phenomenon of resonance.

Timbre and colour of the voice

All the possible ways of modifying the spectral content of a sound determine its timbre proper. However, when referring to the voice, musical language prefers to reserve the term "timbre" for the intrinsic characteristics of a voice, and to use the term "colour" for voluntary modifications of the timbre, mostly for expressive purposes.

There is a simple way to realize that colour is a timbral modification: listening to the same music sung in different languages. The timbre proper, and hence the resulting emotional colour, can undergo even substantial changes.

In the example given here, the famous lullaby by J. Brahms, even a superficial analysis clearly shows the difference at the word Rosen in the original, which falls on the word nanna in the Italian version. The long vowel o in Ro-sen makes it possible to join the two high sounds (E♭₅ and D₅), which form the climax of the first phrase. The peak of greatest intensity in the original version corresponds precisely to the syllable Ro- on E♭₅, followed by a second peak corresponding to en. In the Italian version, the peak that begins to build up on the a of the syllable nan- is immediately interrupted by the double n, which essentially cancels the resonance of the mouth (leaving only that of the nasal cavity), and the sound of the two notes is no longer continuous, even though the singer is performing a "legato". The damping is evident in the time trace (for the spectral analysis see the section on vowels and consonants).

waveform	sonogram	Audio
		VOX_Sample.mp3 VOX_Sample
		VOX_SampleWET.mp3 VOX_SampleWET

Vowels

Some acoustic characteristics of vowels are presented here. See also the pages Simple models for the voice and The vowel A for more details.

In speech

• In speech the vocal folds vibrate as in singing but, since the intonation varies continuously, we are not able to perceive well-defined pitches. However, in some languages, such as Chinese, the intonation of a word determines its meaning and, consequently, the brain is shaped to perceive these small differences too. Moreover, especially in recent studies, the relationship between the evolution of the human species and its ability to recognize sounds has been correlated with the fact that humans seem to be the only species able to develop music, although other species, such as birds, have considerably more developed abilities to produce and recognize sounds.

sounds	pitches	meanings
ma_cinese.mp3 The syllable "ma" pronounced in the four tones of Mandarin Chinese

• When we sing the same note with two different vowels, the harmonic spectrum consists of the same partials. What changes is the distribution of energy over these partials. Looking at the sonogram below, in fact, we can see that the spectral lines corresponding to the harmonic partials are always in the same positions (multiples of 220 Hz, which is the fundamental frequency produced by the singer in this sample). What changes from one vowel to another is the relative intensity of the partials.

sonogram	Audio
	VOX_AEEIOOU.mp3 spoken vowels

• In speech analysis, the difference between vowels is more effectively described in terms of the so-called formants: the resonance frequencies of the vocal tract, which reinforce the nearby harmonics and therefore correspond to the maxima of the envelope of the spectrum. In a vocal tract about 17 cm long there is on average about one every 1000 Hz. Instead of examining each individual harmonic, the first five formants are therefore sought in the spectrum up to 5500 Hz (5000 for the male voice), about one in each 1000 Hz band, and the position of the spectral centroid in each band is studied. This position indicates how the sound energy is distributed in each band and, together with the time variable, is enough to identify the different phonemes in a sonogram unambiguously. The band analysis makes it possible to refer to the same quantities, the formants, regardless of the pitch of the note pronounced. In the figure below we have extracted the formants from the sonogram alongside. Note that the axes, as is the convention in phonology, are swapped: the horizontal axis indicates time, and the vertical axis the frequency divided into formant bands. The following pictures highlight the practical usefulness of reasoning in terms of formants: the spectra of each vowel have too many details to be compared one by one, while the formants give an immediate picture of the differences due to pronunciation alone. Note again that, by convention, the frequency axis is horizontal for spectra, and vertical for formants.

spectra	Formants

	VOX_A.mp3 vowel a		VOX_E.mp3 closed e vowel		VOX_E2.mp3 open e vowel
	VOX_I.mp3 vowel i		VOX_O2.mp3 closed o vowel		VOX_O.mp3 open o vowel
			VOX_U.mp3 vowel u

In singing

Singing allows the expressive extension of the human vocal faculties. Training in singing can take many years, and allows the singer to achieve exceptional performance from many points of view. The intensity of the sounds, their duration, the ability to sing for hours on end without losing the voice, the ability to adjust the colour of the sound to the expressive needs of the score, the ability to optimize the effort and wear of the vocal apparatus, the effective use of all the body's resonances to obtain a pleasant timbre, homogeneous across the range, are just a few examples. For a broad survey of examples, see also the section timbre and positions.

Precisely because of the possibility of varying the colour of the voice, great attention is paid to training pronunciation in singing, in which the number of vowels is generally greater than the canonical five, which in any case may require quite different mouth positions depending on whether they are produced in a low or a high register.

In this section we limit ourselves to considering the relationship between singing and the intelligibility of vowels with a canonical example: the scale from G₄ to G₅ is sung using five canonical vowels. Listening to each scale, one easily recognizes which vowel is used. However, isolating only the highest notes gives a surprise (see below).

waveform	sonogram	Audio
		VOX_ScalaA.mp3 VOX_ScalaA

waveform	sonogram	Audio
		VOX_ScalaE.mp3 VOX_ScalaE

waveform	sonogram	Audio
		VOX_ScalaI.mp3 VOX_ScalaI

waveform	sonogram	Audio
		VOX_ScalaO.mp3 VOX_ScalaO

waveform	sonogram	Audio
		VOX_ScalaU.mp3 VOX_ScalaU

In the following sample we have isolated the five high Gs of the scales above, and put them in a different order. If you listen to the sample, you will no longer be able to tell in which order the vowels are sung! This effect is not due to a mistake in the singing but, on the contrary, to good technique. Indeed, in the high register it is better for the singer to try to concentrate all the sound energy in the fundamental, because the higher harmonics, being damped more efficiently, would contribute little to audibility. Besides this, one must remember that the distinguishability of vowels depends on the density of energy emitted in particular regions of the spectrum (the formants). Moving the fundamental into a higher region, the distances between the harmonics increase, and fewer harmonics fall in the formant region, making the sound less intelligible in any case. The phenomenon is already very evident in this range (mezzo-soprano), and becomes critical for the intelligibility of pronunciation in the very high register of the soprano voice.

All you can do is guess, with a probability of success of 0.83%.^[1]

sonogram	Formants	Audio
		VOX_Vocali_acuto.mp3 VOX_Vocali_acuto

spectra of the five high vowels, averaged over the time of emission
Actually, a careful examination of the average spectrum of each individual high vowel shows minimal differences (a sign that we did not cheat), but the fact is that the information on the quality of the vowels, which is contained within the first five formants below 5500 Hz, is strongly degraded, because the harmonics are more widely spaced. In our case only seven harmonics lie within this band.

Consonants

The case of consonants is different: although they can sometimes be "voiced", in general they have a purely noise-like character (especially the sibilants – "s" and "z" – the labials – "p" and "b" – and the dentals – "t" and "d") and form a "discontinuity" in the course of the intonation of the vowels.

Voiced consonants keep a trace of the partials in harmonic ratios, because they are pronounced together with the voice's own sound. The rolled "r" introduces a rather rapid amplitude modulation (equivalent to a beat frequency of about 28 Hz), while in pronouncing the "n" the resonance of the mouth is strongly damped, or eliminated altogether, in favour of the resonance of the nose. Note, however, that in Italian the letter "n" corresponds to three distinct nasal phonemes: a dental one, a velar one and a labiodental one. For example, pay attention to the position of the mouth while pronouncing the words "analisi", "anche" and "invece". You will notice that in the first case the tip of the tongue touches the front of the palate and the teeth, in the second case it is the soft palate that closes the oral cavity, and not the tongue, and in the third case the lower lip touches the upper incisors, which does not happen in the other two cases. If you try to pronounce "anche" or "invece" using only the dental n, instead of the other versions, you get a kind of "an-e-che" or "in-e-vece", very fashionable in certain singing styles of the 1930s.

dental n	labiodental n	velar n

sonogram	Audio
	VOX_NR.mp3 "n" and "r"

waveforms
N	R

Voiceless consonants have a continuous broadband spectrum, typical of noise. There is a clear difference in the distribution of spectral weight between the fricatives "s" (as in "senno") and "sh" (as in "scena"), and the dental "t".

The "t" is an almost white noise (the spectrum falls off only at very high frequencies). On the other hand, by the classical uncertainty principle, a very short sound (lasting less than 2 hundredths of a second) must have Fourier components spread over the whole spectrum. The "s" is characterized by a single very broad band between 5 and 15 kHz, while the postalveolar sound "sh" is richer in low frequencies and, with its main peak between 2.5 and 8 kHz, corresponds to a warmer coloured noise (see the page sound or noise?, where several examples of noise can be found).

Sonogram	Audio
	VOX_TSSH.mp3 "t", "s", "sh"

waveforms
T	S	SH

Acoustic characteristics

waveform	sonogram	Audio
		VOX_Sib.mp3 VOX_Sib

Timbre and positions

The following examples typically highlight the effect of singing technique. The first vocalise is sung with a "trained" (placed) voice, while the second with a "normal" voice. Voice placement consists in modifying the shape and size of the cavities of the vocal tract so as to enhance some of their natural resonances. Typically these resonances produce an increase in the energy emitted at frequencies well above that of the note being sung.

It is easy to understand that a good singing technique, resulting in a good "tuning" of the resonances of the vocal tract, allows the singer:

1. to generate a sound richer in harmonics (hence with a richer timbre)
2. to generate a more intense sound for the same emission pressure.

In the following two examples a high and a low note (B♭₄ at 458 Hz and B♭₃ at 229 Hz) are sung with the same subjective effort, but the first time with a placed voice, and the second with a natural voice. In both cases one hears that the placed sound is more intense and richer, and in both cases one clearly sees from the sonograms that the effect of placement is almost identical, despite the difference in pitch between the two notes. Placement allows a greater concentration of energy in the band between 2000 and 3000 Hz (third and fourth formants).

sonogram	Audio
	VOX_AcutoPosizione.mp3 VOX_AcutoPosizione
	VOX_GravePosizione.mp3 VOX_GravePosizione

Thanks to training, singers manage to finely adjust the spectral region whose resonance they favour or suppress. This adjustment has two functions: the first is to favour emission at the extremes of the range. The second is to adjust the colour of the notes. One can in fact distinguish between timbre, register and colour. Timbre is an unchangeable attribute of the voice, the quality that lets us distinguish one person's voice from another's. A register is a part of the range produced with the same mode of vibration of the vocal folds: in chest voice the vocal folds vibrate through their whole thickness, in head voice and falsetto mainly their thin edges vibrate^[2]. Colour is the particular shade given to the voice at a certain moment, for expressive purposes, by changing the shape of the vocal tract. The difference becomes clear with the examples.

For teaching simplicity, singers identify three areas, ordered from the darkest to the brightest (only chest voice and head voice are registers in the sense just described; throat voice is rather a way of shaping the vocal tract):

1. chest voice,
2. throat voice,
3. head voice.

Chest voice is used to favour the emission of low notes. Throat voice is mainly adopted in pop music, while head voice enhances the high frequencies.

In the lowest part of the range, chest voice again enhances the third and fourth formants. In the example one hears a G#₃ sung first with a natural voice, and then with chest voice.

waveform	sonogram	Audio
		VOX_TestaPetto.mp3 VOX_TestaPetto

In the following example one hears an F at about 345 Hz sung first deliberately from the throat, then with a placed voice, and finally in head voice, directed "towards the nose". The sonogram clearly shows the different position of the centroid of the spectral intensity in the three cases. The simplified analysis in terms of formants shows the change even better: the first case has "speech-like" formants (see the section on vowels); the placed voice causes the last two formants to crowd together in the region between 3 and 4 kHz; head voice instead causes the suppression of the second formant, in favour of the weight of the third.

sonogram	Formants	Audio
		VOX_GolaVersoIlNaso.mp3 VOX_GolaVersoIlNaso

Effects

It is interesting to note how the voice in piano becomes practically indistinguishable from a pure sound: almost all the energy, in fact, is concentrated in the fundamental. Not by chance, one speaks of a "flute-like" voice. There is a difference of about 21 dB between piano and forte. In piano, out of an instantaneous power of -25.610 dB, -25.617 dB belong to a 200 Hz band centred on the fundamental, with a difference therefore of only 0.007 dB. In forte, instead, these two values are about -3.961 dB and -4.516 dB, with a difference of about 0.5 dB.

waveform	sonogram	Audio
		VOX_PianoForte.mp3 VOX_PianoForte

The dynamic range of the voice (about 21 dB) can also be exploited in "amplitude modulation" to obtain effects such as crescendo (from piano to forte), sforzando (a forte immediately followed by a piano), and vibrato (which is also, and above all, a periodic modulation of the frequency), examples of which follow.

waveform	sonogram	Audio
		VOX_Crescendo.mp3 VOX_Crescendo

waveform	sonogram	Audio
		VOX_Sforzando.mp3 VOX_Sforzando

waveform	sonogram	Audio
		VOX_Vibrato.mp3 VOX_Vibrato

Singing with the mouth closed does not at all mean preventing the emission of sound, so much so that this effect is often used in music of all kinds. There are, however, several ways of obtaining it. The example presents singing with the mouth closed and teeth clenched, with the mouth closed but the teeth apart, and with the mouth open.

waveform	sonogram	Audio
		VOX_BoccaChiusa.mp3 VOX_BoccaChiusa

Besides through voiceless consonants, sounds can be articulated by interrupting the flow of air compressed by the diaphragm, or by small glottal strokes. The sample gives an example of a technique of the first type to obtain the "staccato" effect.

waveform	sonogram	Audio
		VOX_Staccato.mp3 VOX_Staccato

Other effects

There is also a wide range of effects that can be obtained by using the vocal apparatus in a non-standard way.

• In the case of whispered emission, on the other hand, the vocal folds let the air pass without vibrating; the result is a "voiceless" emission, like a breath "filtered" and modulated by the oral cavity alone, which still manages to vary its timbre (so much so that we distinguish different vowels even when whispering) but not to give it the intonation that only the vibration of the vocal folds can produce.

• Falsetto is a particular mode of emission that allows sounds to be produced roughly an octave above the normal speaking register. Although it is generally an innate ability, this technique has been widely exploited in music to allow male singers to perform parts in a female register, or to obtain particular effects. The falsetto voice generally changes its timbre slightly but, if particularly well trained, this technique can also be used to extend the range without the change in timbre being particularly evident. The difference in timbre between a male voice in falsetto, a female voice and a child's voice singing the same notes, on the other hand, generally remains considerable.

Some singers (think for example of Freddie Mercury) have made falsetto a kind of signature of their style. In falsetto the inner muscle of the vocal folds relaxes, while the cricothyroid muscles lengthen and stretch the folds: they become thin and vibrate only with their edges, not with their whole thickness. The vibrating mass is smaller and the tension greater, so the vibration frequency rises^[2][3].

• Subharmonics and double sounds. There are techniques, developed mostly in certain ethnic musics, especially in Asia, that allow a single singer to produce two voices. One "voice" is actually a fixed low sound (drone), while the real melodic voice is obtained by emphasizing the resonances of the oral cavity at the notes corresponding to the melody to be sung. A similar guttural timbre is found in the canto a tenore of Barbagia, in Sardinia, where however the four voices (bassu, contra, boche and mesu boche) are sung by four different singers: it is the two lowest voices, bassu and contra, that have the characteristic deep, guttural timbre^[4].

Sound examples can be found at these links:

• from T. Levin, M. Edgerton, "The Throat Singers of Tuva", Scientific American (September 1999)
• A whole site devoted to the Khoomei style

Check your understanding

Further reading and links

• In the guided path Vowels with TubeLabLight you can change the shape of the vocal tract with the TubeLabLight app and see how the formants of the seven Italian vowels move.
• Related questions and answers:
  • Is it possible to sing two notes simultaneously?
  • Why is it that when a soprano sings in a high register, it is very difficult to distinguish the vowel sounds being sung?
  • Why are we able to distinguish vowel sounds if we pronounce them all with the same intonation?
  • Why do our voices sound so silly after we have inhaled helium from a balloon?
  • Why is it that when I hear my recorded voice it sounds so different from my real voice?
  • Can a saxophonist change the timbre with the mouth and throat?