Cover: AI-generated editorial illustration.
The Open ASR Leaderboard added Egyptian and Nigerian English test splits on November 6, 2026, contributed by Liva AI. Both cover conversational, spontaneous speech with punctuated and cased transcriptions including disfluencies: Egyptian English runs 1.34 hours and Nigerian English 1.55 hours.
The data is held out privately, which matters. Benchmarks that are publicly visible get optimized against directly. Private test sets from independent data providers make it much harder to overfit to the evaluation.
A richer accent breakdown
With these additions, the leaderboard now groups accents using Kachru's Three Circles model. It reports US versus non-US performance, and separately Inner Circle (US, UK, Australia, Canada) versus Outer and Expanding Circles (India, Nigeria, Egypt). As more accents join, those outer groups will be split further.
This framing is worth noting. Most commercial ASR benchmarks remain weighted heavily toward American and British English. A model that scores well on those sets can quietly degrade on accents from the Global South, where real-world usage is growing fast.
Why this matters for developers
If you are building voice products that serve users in Africa or the Middle East, the leaderboard's private tab now offers a more honest signal than top-line word error rate figures. The addition of held-out conversational data, rather than scripted reads, also better reflects how people actually speak in those markets. The leaderboard is a free public resource maintained on Hugging Face.