live-coding-academy_ ⌕ search
home/ atoms/ formant-synthesis

Voice synthesis models the vocal tract as a series of formant resonances shaped by the source-filter model

Concept L3Craft B · Sound design & synthesisE · Modular, Eurorack, grooveboxes & dawless

The human voice is well-described by a source-filter model: a glottal source (periodic for voiced sounds, aperiodic for unvoiced) excites a resonating vocal tract. The key perceptual features of speech sounds are the formant frequencies (F1 to F5): major spectral peaks caused by resonances of the vocal tract cavities (larynx, pharynx, oral cavity, nasal cavity). Five formant positions can be enough to describe most vowel sounds. Voice synthesis using formant synthesis involves generating source waveforms and passing them through bandpass filters tuned to the target formant frequencies. This principle underlies cascade and parallel formant synthesizers and text-to-speech systems.

Examples

Kelly and Lochbaum’s vocal tract simulation (1962) at Bell Labs. The DECtalk synthesizer used formant synthesis. Contemporary TTS systems like Tacotron have largely superseded formant methods.

Assessment

What are vocal tract formants and why do five of them suffice to distinguish most vowel sounds? How does the source-filter model separate the pitch of the voice from its vowel quality?

Prerequisites — understand these first

Don't confuse with

Sources

“of larynx, pharynx, oral cavity (including tongue and teeth), nasal cavity and lip. Most important in the perception of speech are formant regions in the spectrum, which are major spectral peaks”
corpus · nick-collins-introduction-to-computer-music-free-author-edit · chunk 69

Taught in modules

Unlocks (2) — atoms that build on this