How a phase vocoder separates pitch from timing
A recording is a changing mixture of sound frequencies. A phase vocoder analyzes small, overlapping slices of that sound, tracking the strength and phase of each frequency component. Phase describes where a repeating wave is in its cycle; keeping track of it helps the software rebuild the slices without clicks or gaps.
To raise or lower a singer’s pitch, the software shifts the frequency components and synthesizes them again. But simply shifting every frequency would also speed up or slow down the recording, like playing a tape at a different speed. A phase vocoder avoids that by treating frequency and duration as separate controls: it changes the frequencies while placing the analyzed slices back across the original timeline.
Why the result can still sound different
Keeping the track’s length does not guarantee a perfect, natural-sounding voice. The process can blur sharp attacks—such as consonants—or create a slightly watery, smeared quality, especially with large pitch changes. That is because the software must estimate and reshape sound within short slices, rather than moving a single, perfectly isolated pitch knob.
More advanced processing can preserve parts of the sound that make a voice recognizable, such as its formants—the broad resonances that give a voice its characteristic tone. The central tradeoff remains: the singer can sound higher or lower without the beat and phrase lengths shifting, though the timbre may change along the way.
