phasevocoder

A phase vocoder is a class of digital signal processing algorithms used for time-scale modification and pitch shifting of audio signals. It works by transforming the input signal into the frequency domain with a short-time Fourier transform (STFT), which yields frames that contain both magnitude and phase information for the spectral components. The frames are analyzed with a chosen window size and overlap, and the magnitude and phase are tracked across successive frames.

In operation, the phase vocoder analyzes consecutive frames to determine their instantaneous frequency from phase differences and unwraps the phase to maintain continuity. The spectral information can then be modified in the frequency domain: for time-scale modification, the synthesis hop size is changed relative to the analysis hop size, effectively stretching or compressing the signal in time while attempting to preserve pitch; for pitch shifting, spectral content can be adjusted and later resampled in time to achieve the desired pitch without changing duration. The modified spectra are then converted back to the time domain via an inverse STFT, and the frames are recombined using overlap-add to produce the output signal.

Classical phase vocoders excel at smooth, continuous changes but can introduce artifacts around transients, leading to smearing of sharp attacks. Various enhancements address these issues, including phase locking or transient-preserving techniques that group spectral bins and maintain phase relationships within bands, and methods such as SOLA-based approaches to improve transient handling.

Applications include audio time-stretching for music production and speech processing, automatic tuning and vocal effects, sound design, and any scenario requiring controlled modification of duration or pitch without drastic resynthesis.