mm2002-pn/whisper-wolof
Domain:
natural language processingRecord type:
modeldataset
Creator:
mm2
Host:
# Whisper Wolof - Fine-tuning ASR
Fine-tuning OpenAI Whisper pour la reconnaissance de la parole en Wolof.
## Modèles disponibles
- **Phase 1** (2000 fichiers, 2 epochs) → mansourr/wolof-asr-commonvoice
## Dataset
- 44,000+ fichiers audio Wolof (CommonVoice)
- HuggingFace: mansourr/wolof-asr-commonvoice
## Usage
```bash
pip install transformers librosa torch
python3 train_whisper_phase2.py
```
## Résultats
| Modèle | Fichiers | WER | CER |
|--------|----------|-----|-----|
| Phase 1 | 2000 | 65.0% | 30.1% |
| Merged (P1+P2+P3) | 6000 | 65.6% | 27.0% |