Formant Synthesis

Replace the filter with a few resonant peaks and a synth can sing vowels.

Related

TimbreSynthesizerThe vocoderContemporary

The question

A singer's pitch barely changes between an ah and an oo; the difference lies in the mouth's resonances. Can a synth make vowels from a few peaks without recording a voice?

How it works

Vowel energy gathers around a few formants. The first sits near 300 to 800 Hz, the second near 800 to 2500 Hz. Formant synthesis does not mimic the vocal folds; it feeds a harmonic source into narrow resonators, and peak positions decide which vowel is heard. Narrow, tall peaks make the vowel clear; broad peaks turn it into a vague hum. A vocoder uses a real voice to control a filter bank, but both depend on formants.

Settings and what you hear

Lifting the first peak makes the tone throatier; moving the second forward pulls an ah toward an eh. Peak bandwidth sets intelligibility: narrow reads as a machine speaking, wide as singing behind a hand. A little breath noise adds humanity, but too much masks the pitch. When a glide moves between vowels, the peaks must travel too or the sound breaks.

What to listen for

Hear a choir pad and notice that it seems to hold words without saying them. Then hear a whistle-like tone, where a narrow peak makes the top sound like oo. Finally compare a plain sawtooth; it has no peaks, so it is only bright and never seems to sing.

Common misconceptions

That formant synthesis can only imitate voices; that more peaks sound more human, when the first two matter most; that it is the same thing as a vocoder.

Sources

• Miller Puckette, The Theory and Technique of Electronic Music — formant synthesis checked, not quoted
• Curtis Roads, The Computer Music Tutorial — vocal synthesis checked, not quoted

Used for fact-checking only; all prose is original.