Are formants measured in hertz?

Formant center frequencies and bandwidths are commonly expressed in hertz, while other associated properties use different units.

Can formants change during one sound?

Yes. Vocal-tract movement can produce formant trajectories, including the changing pattern within a diphthong.

Does a software label guarantee a true vocal-tract resonance?

No. Model-fitting artifacts can appear as estimated formants, so the output needs interpretation and quality checks.

AI Glossary / Natural language and speech / Intermediate

Formant

Definition

A formant is a resonance-related concentration of energy in the speech spectrum, commonly described through the frequency and bandwidth of a vocal-tract resonance. Formants help characterize speech sounds, especially vowels. The lowest formant frequencies are conventionally labeled F1, F2, F3, and so on in ascending order.

What it means

The vocal tract shapes sound rather than simply passing it through unchanged. Its resonances emphasize some frequency regions more than others. Changes in tongue position, mouth opening, and lip configuration can therefore alter the spectral pattern.
Formant frequencies are different from the fundamental frequency, F0, which is associated with the repetition rate of voiced sound and perceived pitch. A speaker can change pitch while broadly maintaining a vowel quality, or change vowel quality without making the same change in pitch.
The term is used with slightly different emphases in acoustic descriptions: a vocal-tract resonance, its estimated frequency, or the resulting spectral prominence. An observed spectral peak and a physical resonance are related but should not always be treated as exactly identical measurements.

How it works

In a simplified source-filter view, a sound source supplies acoustic energy and the vocal tract filters it. The filter’s resonances influence which parts of the source spectrum are emphasized.
An analysis estimates formant frequencies and bandwidths over short intervals. A common approach uses linear predictive analysis to model the spectral shape. Estimates can then be plotted as changing tracks through an utterance.
For vowels, F1 and F2 are especially useful descriptive dimensions. Broadly, vowel height relates inversely to F1, while front - back distinctions are reflected in F2, with important effects from rounding and the speaker’s vocal tract.
Automatic tracks are estimates, not unquestionable observations. Analysis settings, recording quality, the sound being produced, and the spacing of source harmonics can affect interpretation.

Three practical examples

A phonetics researcher measures F1 and F2 to compare vowel realizations within a speech sample.
A speech-synthesis experiment changes resonance settings while holding the source pitch constant to study their separate effects.
An annotation team checks suspicious formant tracks against the spectrogram and recording before including measurements in a dataset.

Think of it this way

Think of the same buzzing source passing through differently shaped resonating spaces. The source can keep its rhythm while the spectral emphasis changes. The analogy is limited because a human vocal tract is dynamic and interacts with the source; it is not simply a fixed collection of independent musical resonators.

Worked example

Consider an invented analysis frame with source fundamental frequency F0 = 100 Hz and estimated formants F1 = 500 Hz and F2 = 1500 Hz.
For a periodic source at 100 Hz, harmonics occur at 100, 200, 300 Hz, and so on. The fifth harmonic is at 500 Hz and the fifteenth at 1500 Hz. In this simplified case, those harmonic frequencies coincide with the specified resonance centers.
Now change the source to F0 = 120 Hz while keeping the model’s resonance centers at 500 and 1500 Hz. The nearby harmonics are now 480 and 600 Hz around F1, and 1440 and 1560 Hz around F2. The harmonic spacing changed, but the stated filter resonances did not.
This illustrates why formants are not simply fixed-number harmonics. A strong observed harmonic near a resonance need not land exactly at its center.
The values are teaching parameters, not universal measurements for a named vowel or speaker. Real estimation also requires bandwidths, an appropriate analysis interval, and checks that the fitted model represents the recording sensibly.

Illustrative case study

In a fictional project, Hazel Phonetics compares vowel recordings from a small teaching dataset. An automatic analysis occasionally displays a sudden jump in F2 during an otherwise steady vowel.
The analyst reviews the audio and spectrogram and tests whether the chosen analysis range is suitable. Questionable frames are marked for review instead of being interpreted immediately as a real articulatory change.
The imagined case shows why quality control matters for acoustic measurements. It does not imply that every jump is an error or that one set of software defaults is appropriate for every speaker and vowel.

Common misconceptions and mistakes

F1 is not F0. The first formant describes a resonance-related feature, whereas F0 describes the periodic source’s repetition rate.
Formants are not the same as individual harmonics. Harmonics are spaced according to the source frequency; resonance regions shape their relative amplitudes.
A formant number is also not a universal vowel label. Speakers and contexts vary, and interpretation often depends on several measurements together.
Finally, a smooth-looking automated track can still be wrong. A visually tidy output does not remove the need to inspect the signal and analysis assumptions.

Business use

Formant analysis supports speech research, pronunciation resources, acoustic modeling, and controlled synthesis experiments. It provides interpretable features that can help explain differences between sounds.
For practical datasets, document the analysis method, settings, measurement interval, and quality-review process. Comparisons across speakers may require careful normalization or modeling rather than direct comparison of raw frequencies alone.
Avoid turning one estimated formant value into an unsupported personal assessment. The useful result is an appropriately qualified acoustic description, grounded in the recording and the purpose of the analysis.

Check your understanding

If F0 changes from 100 to 200 Hz while a synthetic filter keeps F1 at 500 Hz, must F1 become 1000 Hz?

Answer: No. The source’s harmonic spacing has changed, but the model’s first resonance remains at 500 Hz. Source and filter parameters are distinct in this example.

Related terms

Learn next

Learn the difference between a waveform, a spectrum, and a spectrogram. Then compare F0 with F1 and F2 in a carefully chosen vowel recording. Read the formant-analysis settings for your software and inspect difficult cases before treating automatically produced tracks as measurements ready for comparison.

Sources

Praat, official source-filter tutorial: https://www.fon.hum.uva.nl/praat/manual/Source-filter_synthesis_2__Filtering_a_source.html. Praat, official Burg formant-analysis documentation and limitations: https://www.fon.hum.uva.nl/praat/manual/Sound__To_Formant__burg____.html. Penelope Eckert, author-hosted vowel explanation and research context: https://web.stanford.edu/~eckert/vowels. Independent review checked Praat’s source-filter and Burg-analysis documentation, including artifacts and speaker-dependent settings, and Eckert’s formant explanation.

Previous: Formal logic Next: Formant synthesis