metadata
license: cc-by-nc-4.0
Model Checkpoint for LauraTSE.
We have trained it on LibriSpeech using Dynamic Mixing with SNR 0-5 dB for 100 epochs and then finetune it on Libri2Mix for 20 epochs.
Results on Libri2Mix clean testset:
| Model | SIG | BAK | OVRL | NISQA | SpeechBERT | dWER | WavLM Sim | Wespeaker Sim |
|---|---|---|---|---|---|---|---|---|
| LauraTSE | 3.609 | 4.084 | 3.336 | 4.333 | 0.908 | 0.159 | 0.974 | 0.876 |
Please refer to files for model ckpt and config.