mmmt-tools / Synthesis / FOF formants
on-device · nothing leaves this page

FOF — granular formant synthesis

The other way to make a voice. Where the source–filter model filters a buzz, FOF (Xavier Rodet's Fonction d'Onde Formantique, the engine inside CHANT) builds each formant out of grains: a tiny damped sinusoid at the formant frequency, fired once every vocal-fold period. The grain's decay sets the formant's width; its soft attack sets the skirt. Overlap the grains from successive periods and the formant appears — pitch-synchronous granular synthesis, and the classic way to synthesize a convincing sung voice. Related to VOSIM, which builds formants from pulse trains instead.

waveform (grains)
spectrum (formant peaks)

Vowel

0.00

Grain

3.0 ms
0.30 s

A shorter attack widens each formant's skirt (brighter, buzzier); a longer one narrows it (purer, more hollow).

sing a note (a s d f …) — each keypress is a stream of formant grains