LibriSpeech MMS Q-Former Aya
This checkpoint was trained on the full 960-hour LibriSpeech training set. During training, the MMS-300M speech encoder and Q-Former were trainable, while the Aya decoder remained frozen.
Evaluation
The model was evaluated on the LibriSpeech test-clean split.
| Metric | Score |
|---|---|
| Character error rate (CER) | 2.80% |
| Word error rate (WER) | 6.78% |