Records the input into a circular buffer and slices it at zero crossings. A group of wavesets forms a collection; collections are played back reordered, with dropouts and repeats. Timing is driven by the source material rather than by tempo, so identical settings sound different on different material.
If you get no sound, serve the page over a local server:
python3 -m http.server. Audio worklets usually refuse to load
straight from the file system.