w2v2-bert Wolof 75h
Wolof ASR, MIT
Official site consultedAI: models
Last verified: 1 October 2026
Wolof speech recognition model of about 600 million parameters, fine-tuned from w2v-bert-2.0 on 75 hours of mixed Wolof speech. It shows a word error rate of 0.4378 on the evaluation set, after 50 training epochs. MIT license.
Services
- Wolof speech recognition
Access requirements
Open weights on Hugging Face, MIT license.
Sources
- w2v2-bert Wolof 75h (new tab), verified on 1 October 2026Official site consulted