When a model processes audio, a long signal is split into fixed-length chunks. Overlap is the degree of overlap between adjacent windows. 

For example:
overlap = 0% → windows are placed back-to-back (no overlap)
overlap = 50% → each subsequent window overlaps the previous one by half
overlap = 75% → 75% overlap, and so on

Overlap helps smooth transitions between windows, use more context, and improve separation quality. Most models on the site use 50% overlap unless stated otherwise. A more visual explanation is shown in the illustration below: