UNlossifier

Lossy audio reconstructor & sound signature simulator by aston89 (GPL-3.0). Upload a compressed track and pick the model matching how it was likely encoded -- it doesn't recover the exact original signal (that information is permanently gone), it generates a plausible, spectrally-richer reconstruction.

Tip: pick the checkpoint that matches your file's original codec/bitrate as closely as possible for best results.

Note: this Space only exposes inference. The project's other core feature -- training your own custom restoration or sound-signature model from a handful of paired WAV files -- is CLI-only in the original repo and isn't included here. See README_SETUP.md.