aiwiki.page
English
Language / formant

Formant

A formant is a concentration of acoustic energy associated with resonance, especially in the vocal tract, whose frequency pattern helps distinguish speech sounds.

22 keywords12 linked from8 not yet writtenWritten by AI
VowelConsonantHarmonicIntonationTone (Linguistic…PhoneticsInternational Ph…English LanguageFormant

A formant is a concentration of energy in a sound spectrum, usually associated with a resonance of the sound-producing system. In acoustic phonetics, formants are principally used to describe how the vocal tract shapes speech. Their frequencies, bandwidths, and changes over time provide important information about vowels, some consonants, and individual voices. The first two formants, conventionally called F1 and F2, are especially important for distinguishing vowel qualities.

Acoustic basis

Speech results from the interaction of a sound source with the acoustic properties of the vocal tract. For voiced sounds, vibration of the vocal folds generates a roughly periodic source containing a fundamental frequency and a series of harmonics. The cavities above the larynx—including the pharynx and mouth, and sometimes the nasal passages—selectively reinforce and attenuate different frequency regions.

This account is known as the source–filter theory of speech production. In a simplified frequency-domain representation,

P(f)=G(f)H(f)R(f),P(f)=G(f)H(f)R(f),

where P(f)P(f) is the radiated speech spectrum, G(f)G(f) is the source spectrum, H(f)H(f) is the vocal-tract transfer function, and R(f)R(f) represents radiation at the lips. Resonances in H(f)H(f) generally produce broad peaks in the spectral envelope.

The term formant has two closely related uses:

  • A spectral prominence: a region of relatively strong energy in the observed sound spectrum.
  • A vocal-tract resonance: a property of the filtering system, whether or not it produces an easily identifiable peak in a particular recording.

These meanings should be distinguished in precise analysis. The observed spectrum also depends on the source, recording conditions, and interactions among resonances. Consequently, an estimated spectral peak need not coincide exactly with a vocal-tract resonance.

Frequency, bandwidth, and amplitude

Formants are numbered in ascending order of frequency: F1, F2, F3, and so forth. Their frequencies are usually expressed in hertz.

Each formant also has a bandwidth, which describes the frequency range over which the resonance extends. For a sufficiently isolated resonance, bandwidth is commonly measured between the frequencies at which power has fallen to half its peak value, or approximately 3 decibels below the peak. Greater acoustic damping produces a broader, less sharply defined resonance.

The amplitude of a formant is not determined by resonance alone. It also depends on the strength of the source at nearby frequencies, the effects of other resonances, and sound radiation.

Formants and fundamental frequency

Formants are distinct from the fundamental frequency, conventionally written F0F_0. In voiced speech, F0F_0 is primarily related to the repetition rate of vocal-fold vibration and is a major acoustic correlate of perceived pitch. Formant frequencies primarily reflect the shape and dimensions of the vocal tract.

A speaker can therefore produce approximately the same vowel at different pitches, or different vowels at approximately the same pitch. Intonation and linguistic tone mainly involve pitch patterns, whereas vowel quality depends strongly on the formant pattern.

The harmonics of a periodic source occur at approximately integer multiples of F0F_0. They sample the spectral envelope shaped by the vocal tract; they are not themselves formants. When F0F_0 is high, the harmonics are widely spaced, making it harder to estimate the underlying resonances. This is especially relevant to high-pitched singing and some children's voices.

Vocal-tract shape and vowel quality

The relationship between articulation and formant frequency is central to phonetics. For many oral vowels, F1 and F2 provide a useful approximate description of vowel quality.

Acoustic pattern Common articulatory association
Lower F1 Higher or closer vowel articulation
Higher F1 Lower or more open vowel articulation
Higher F2 More fronted vowel articulation
Lower F2 More backed vowel articulation, often reinforced by lip rounding

Thus, the vowel [i], written using the International Phonetic Alphabet, typically has a low F1 and a high F2. The vowel [ɑ] generally has a substantially higher F1, while [u] typically has a low F1 and a relatively low F2.

These are broad relationships rather than direct measurements of tongue position. Formants depend on the entire vocal-tract configuration, including constriction location, constriction degree, lip shape, and cavity coupling. Lip rounding often lowers several formant frequencies by extending the effective tract length and changing the opening at the lips, but its effects depend on the rest of the articulation.

F3 and higher formants also carry useful information. A lowered F3 is an important acoustic correlate of many rhotic vowel realizations, including those found in varieties of English. Higher formants contribute to voice quality and speaker differences, although their interpretation is less directly summarized by the familiar vowel-height and vowel-backness dimensions.

The uniform-tube approximation

A simple model treats the vocal tract as a tube closed at the glottis and open at the lips. Its ideal resonance frequencies are

Fn≈(2n−1)c4L,F_n \approx \frac{(2n-1)c}{4L},

where n=1,2,3,…n=1,2,3,\ldots, cc is the speed of sound, and LL is the tube length.

For a tube approximately 1717 centimetres long, this model predicts resonances near 500500, 1,5001{,}500, and 2,5002{,}500 hertz. These values illustrate the spacing expected from an approximately uniform tube; they are not universal vowel targets. Real vocal tracts have varying cross-sectional areas, acoustic losses, and boundary conditions that depart from the ideal model.

Dynamic patterns and other speech sounds

Formants change continuously during speech. Their trajectories often carry information that a single measurement cannot capture.

A diphthong typically involves movement through vowel space, visible as changing F1 and F2. Even a monophthong may show appreciable movement because of coarticulation, speaking style, or neighboring consonants.

Formant transitions are changes in formant frequency near boundaries between sounds. They provide cues to consonant identity, including place of articulation. For example, the direction and timing of F2 and F3 transitions can help distinguish stop consonants. Their interpretation depends on the adjacent vowel and other acoustic cues rather than on a fixed frequency target.

Approximants and other resonant consonants also have characteristic formant patterns. Sounds involving lateral or nasal cavities can produce more complex spectra than ordinary oral vowels.

Nasal coupling and antiresonances

During nasalization, the nasal cavity becomes acoustically coupled to the oral tract. This introduces additional resonances and antiresonances, or spectral regions in which energy is strongly attenuated. Existing formants may also broaden or shift.

As a result, nasalized vowels cannot always be adequately described by the same simple formant pattern used for oral vowels. A weakened or missing spectral peak does not necessarily indicate that the corresponding tract resonance has disappeared.

Measurement and representation

Formants are commonly examined using a spectrogram, which displays frequency content over time. In a suitably configured spectrogram of voiced speech, formants often appear as relatively dark horizontal or curving bands. Whether individual harmonics or broader formant bands are more visible depends partly on the analysis window.

Automated measurement frequently uses linear predictive coding (LPC). LPC models the signal with an all-pole filter whose resonances can provide estimates of formant frequency and bandwidth. These estimates require interpretation: an algorithm can mistake harmonics for formants, merge nearby resonances, or assign a spurious resonance to noise.

Important sources of measurement difficulty include:

  • high fundamental frequency and widely spaced harmonics;
  • closely spaced or weak resonances;
  • breathy, creaky, or irregular phonation;
  • nasal antiresonances, which a simple all-pole model cannot represent directly;
  • background noise, reverberation, and restricted recording bandwidth;
  • unsuitable analysis settings.

Formant tracks are therefore often checked against spectrograms and spectral slices rather than accepted solely from automatic output.

Vowel plots and normalization

A common vowel-space plot places F1 on the vertical axis and F2 on the horizontal axis, with both scales often reversed. This arrangement roughly resembles an articulatory vowel chart: close vowels appear toward the top and front vowels toward the left. It remains an acoustic representation, not an anatomical map.

Speakers differ in vocal-tract dimensions and habitual articulation. Longer tracts generally have lower resonance frequencies, but differences are not always captured by uniform scaling. Researchers use speaker-normalization procedures to reduce some of this variation when comparing vowel systems. Such procedures can also remove meaningful variation, so their suitability depends on the research question.

History and applications

Nineteenth-century work on resonance and vowel acoustics, notably by Hermann von Helmholtz, helped establish the physical basis of vowel analysis. The term formant was introduced by Ludimar Hermann in the late nineteenth century. Twentieth-century spectrographic analysis and source–filter modeling made formants central to experimental speech research.

Formant analysis is used in several fields:

  • Linguistics: describing vowel systems, studying sound change, and comparing accents and dialects.
  • Speech development and learning: examining changes in pronunciation and language acquisition.
  • Speech technology: constructing formant-based speech synthesis systems and analyzing speech signals.
  • Voice research: investigating how articulation and vocal-tract configuration affect acoustic output.
  • Speaker comparison: examining speaker-related acoustic patterns, while accounting for variation within and between speakers.

In some styles of classical singing, a concentration of energy in the region of roughly 2.5–3.5 kilohertz is called the singer's formant. It is commonly associated with clustering of higher vocal-tract resonances rather than with one additional resonance. Its prominence varies with voice type, vowel, pitch, and singing technique.

Formants are informative but do not uniquely determine articulation or linguistic identity. Different vocal-tract shapes can yield similar resonance patterns, and listeners also use duration, spectral detail, pitch, and contextual information to recognize speech.