TODO
Planned work for this repository.
- Train the adapter on the prepared synthetic-room corpus.
- Track reconstruction quality at each saved checkpoint via a held-out validation set of unseen rooms.
- Report the final reverb-quality benchmark on a held-out test split (spectrogram distance and reverberation-time error after decoding).
- Publish the inverse-bijection decoder and the schematic renderer used at training time.
- Release weights, decoder, renderer, and a representative set of demonstration (room schematic, predicted impulse response, dry-signal-with-recovered-reverb) bundles accompanying the model card.