WSOLA

WSOLA, or Waveform Similarity Overlap-Add, is a time-scale modification technique designed to change the duration of audio signals—most notably speech—without significantly altering pitch. Introduced by Steven Verhelst and Dirk Roelands in the early 1990s, WSOLA is widely used in applications that require tempo changes such as playback speed adjustments, teleconferencing, and speech synthesis.

The method operates by dividing the input signal into short overlapping frames and reconstructing the output signal by overlapped addition. A key feature is a search for the best alignment between consecutive frames within a defined window. Specifically, WSOLA evaluates the similarity between the current frame and potential overlap positions, typically using a cross-correlation or error criterion, and selects the position that yields the most coherent transition. The chosen alignment minimizes audible discontinuities at frame boundaries. By adjusting the rate at which input samples feed the output (the time-scale factor), WSOLA can slow down or speed up the audio while preserving much of the original spectral characteristics.

WSOLA is known for producing natural-sounding speech compared with naive time-stretching methods and many earlier overlap-add approaches. It is relatively computationally efficient, making it suitable for real-time processing. Limitations include reduced performance on music or highly dynamic signals and potential artifacts under extreme time-scale factors or reverberant conditions. It remains one of several standard techniques for audio time-scale modification, alongside PSOLA and SOLA.