The phase vocoder converts audio to a time-varying spectrum enabling pitch-time manipulation

The phase vocoder (PV) uses the STFT to convert an audio signal into a complex Fourier representation of time-varying amplitude and phase for each frequency bin. Phase differences between successive frames determine each bin’s true instantaneous frequency (not just its bin center). This richer representation enables high-quality transformations: pitch shifting (scale frequencies of resynthesis components), time-stretching (move frame onset times apart), and pitch-time changing (combine both). The PV can be resynthesized with overlap-add (efficient) or oscillator bank (more robust for large transformations). Key parameters: window size, FFT size, hop size, and window type. The PV produces characteristic artifacts on noisy or transient signals.

Examples

Phase vocoder time-stretch to 2x: analyze with overlap-add PV, then resynthesize with frames spaced 2x farther apart. Pitch unchanged. Pitch-shift up an octave: multiply all resynthesis frequencies by 2.

Assessment

What advantage does the phase vocoder have over simple time-domain granular pitch-time changing? What is the characteristic artifact sound of phase vocoder processing on transient signals?