Repairs speech recordings damaged by clippings. A
generative model reconstructs the samples that were destroyed.
1git clone https://github.com/tdstt22/audio-restore
2cd audio-restore && uv sync
3python restore.py recording.wav restored.wav
The model works on spectrograms at 24 kHz mono. It is given the damaged audio
and a mask marking which samples the clipping destroyed, so it never has to
guess where the damage is. Rather than emitting audio directly, it predicts the
direction from noise toward clean speech, and sampling follows that direction
over to arrive at the reconstruction.
