8 min read
What we found testing Whisper-1 on Wolof in 2026
OpenAI's whisper-1 posts the highest mean WER (1.05) of any system in our 104-sample Senegalese voice benchmark. Here is what that number actually looks like in production.
Read →Field notes
State of African voice AI.
Monthly notes on what specific models do on specific Wolof and French inputs — and what product teams should plan for.
New post roughly once a month. For the underlying benchmark, see the State of Wolof Voice AI 2026 report.