M-Kiriku-ASR
Wolof, Pulaar, Serer ASR
Official site consultedAI: models
Last verified: 1 October 2026
Speech recognition model for Wolof, Pulaar and Sereer, fine-tuned from Whisper Large-v3 on about 150 hours of speech, including 75.4 hours from the Kallaama project. Word error rates: 16.34% Wolof, 45.28% Pulaar, 42.99% Sereer. Apache 2.0 license.
Services
- Wolof, Pulaar and Sereer speech recognition
- Custom language tokens (wo, pu, se)
Access requirements
Access with a Hugging Face login and sharing of contact information; Apache 2.0 license.
Sources
- M-Kiriku-ASR (new tab), verified on 1 October 2026Official site consulted