Egyptian and Nigerian English join the Open ASR Leaderboard

The Open ASR Leaderboard, which ranks speech-to-text models, has added Egyptian and Nigerian English to its private evaluation data through a new partnership with Liva AI, Hugging Face announced in a blog post dated November 6, 2026. The post is credited to Eric Bezzam of Hugging Face along with Ashley Mo and Aoi Otani of Liva AI. Each accent gets one new held-out split of spontaneous conversational speech, about 1.34 hours for Egyptian and 1.55 hours for Nigerian, with transcripts that keep punctuation, casing and disfluencies.

The additions build on private data introduced a few months earlier: 11 splits from Appen and DataOceanAI spanning scripted and conversational speech in five accents, followed by an Indian-accented English split from Voice Arena. The authors frame the expansion as serving two goals, broader representation in speech recognition testing and greater trust in results, with fresh private data and new suppliers intended to limit what they call "benchmaxxing," or tuning models to known tests.

The leaderboard's Private tab will also show a more detailed accent breakdown. Results will compare US against non-US English, and separately compare Inner Circle accents (US, UK, Australia, Canada) with Outer and Expanding Circle accents (India, Nigeria, Egypt), a grouping drawn from Kachru's Three Circles model and credited to an idea from Hume AI's leaderboard. The team says it will separate the Outer and Expanding Circles as more accents arrive.

Liva AI describes itself as gathering multimodal interaction data through consumer products used by real people, which it says yields more natural behavior than task-based collection.