Formants: why shifted voices chipmunk, and how they don't

Play a recording of a voice at double speed and you get the chipmunk: higher pitched, yes, but also smaller, thinner, wrong. Understanding why is the key to understanding every serious voice tool.

A voice is a source and a filter

Voiced sound is made in two places at once. The vocal folds produce the source: a steady train of tiny pulses, and the rate of those pulses is the pitch. Everything above them (throat, mouth, nasal cavity) acts as a filter over that pulse train. The filter has resonant peaks, and those peaks are the formants.

Formants are where most of a voice's identity lives. They make an ee different from an ah, and they make your ah different from someone else's: the filter is literally the shape and size of your vocal tract. Crucially, when you sing a higher note, your pulse rate goes up but your throat stays the same size: pitch moves, formants don't.

The chipmunk, explained

Speeding up a recording (or naively resampling it) scales everything: pulse rate and filter together. The pitch rises, and the formants rise with it, which the ear reads as the vocal tract shrinking. Higher formants mean a smaller instrument; a smaller instrument means a smaller creature. Hence the chipmunk going up, and the sludgy giant coming down. The voice hasn't just changed pitch; it has changed species.

Formant-preserving shifting

The fix is to treat the two layers separately: change the pulse rate, keep the filter where it was. Formant-preserving pitch shifters do exactly this: time-domain methods rebuild the voice pulse by pulse and counter-scale each pulse's contents; spectral methods estimate the filter's envelope and re-pin it after shifting. Either way, the result is the same person singing a different note, rather than a different person.

This is also why voice tools expose a formant control at all: once pitch and formants are decoupled, the formant scale becomes a free, creative parameter. Keep it centred and a shifted voice keeps its own throat; push it deliberately and you resize the singer without touching the note.