The phase vocoder enables independent time-stretching and pitch-shifting by operating on FFT analysis frames

The phase vocoder (PV) analyses a signal frame-by-frame using short-time FFTs, tracking the amplitude and phase of each frequency bin over time. This representation separates the signal into slowly-varying frequency and amplitude trajectories. Time-stretching is achieved by computing more (or fewer) synthesis frames from the same analysis frames—stretching or compressing the time axis without changing spectral content. Pitch-shifting is achieved by scaling all frequency bin positions up or down before resynthesis. Unlike simple tape speed changes, which alter both pitch and duration together, the PV can change each independently. The phase coherence between successive frames is critical; phase unwrapping preserves the spectral structure of transients during transformation.

Examples

A vocal sample at 120 BPM can be time-stretched to 80 BPM (0.67× speed) without lowering pitch, using PV analysis hop size of 256 samples and synthesis hop of 384 samples at 44.1 kHz. Autotune-style pitch correction applies PV pitch-shifting while preserving duration.

Assessment

Explain why a phase vocoder can pitch-shift a voice without changing its duration, while playing back a tape recording at a different speed cannot achieve this. Identify the step in the PV algorithm that controls each parameter independently.