YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Speech enhancement + ASR checkpoints
Trained model weights (PyTorch .pt) for generative speech enhancement with
discrete speech tokens, built on Qwen2.5-0.5B and Whisper.
38 checkpoints, one per training run.
Method, training details and results are described in the accompanying paper (in preparation). This repository contains weights only — no code and no data.
Base models
- Qwen2.5-0.5B — Apache-2.0
- Whisper — MIT
- CosyVoice speech tokenizer
Notes
Checkpoints are torch.save state dicts and are not directly loadable with
transformers; they require the training code's model definition.
Some runs were trained on corpora with restrictive licenses (LDC). Only model weights are distributed here, never any corpus data or derivatives of it.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support