There’s a new demo page for v3 of the iterative audio decomposition model I’ve been working on.
The model is small and hasn’t had much (less than 24-hours) of training time so far, but doubles the duration of audio segments from the prior model.

You can read more and listen to demos here