timestretching

Timestretching, or time-scale modification (TSM), is the process of changing the duration of an audio signal without significantly altering its pitch. It allows speeding up or slowing down playback while maintaining the original tonal characteristics. Pitch shifting, by contrast, changes pitch with or without duration changes, depending on the method used.

The most common approaches to timestretching fall into two broad categories: frequency-domain methods and time-domain methods. Frequency-domain methods, such as the phase vocoder, analyze short-time windows with a Fourier transform, adjust the timeline by resampling in the spectral domain, and reconstruct the signal with phase adjustments. They can preserve pitch well for steady tones but may smear transients or introduce metallic artifacts if not carefully configured. Time-domain methods, including SOLA (synchronized overlap-add), WSOLA (waveform similarity overlap-add), and PSOLA (pitch-synchronous overlap-add), operate by segmenting the signal and overlapping or splicing segments to change duration. These methods often yield more natural-sounding results for speech and musical material with clear transient structure, but rely on reliable segmentation and, in some cases, accurate pitch estimates.

Quality and artifact considerations include transient smearing, phasiness, warbling, and timbre alteration. Real-time timestretching requires efficient algorithms and buffering strategies; offline processing can achieve higher quality with more computation. Variants and hybrids aim to balance artifacts across different audio content.

Applications span music production, film and video post-production, broadcasting, game audio, and speech processing, where changing tempo without affecting pitch is desirable for synchronization, practice, or playback speed adjustments.