Hugging Face adds Egyptian and Nigerian English accents to Open ASR Leaderboard with Liva AI data
Two new conversational speech datasets expand accent coverage and introduce private evaluation splits to reduce benchmark gaming.
1 source · cross-referenced
- Two new conversational speech datasets for Egyptian and Nigerian English have been added to the Open ASR Leaderboard.
- The datasets, provided by Liva AI, total 1.34 hours and 1.55 hours respectively and include spontaneous, punctuated, cased transcriptions with disfluencies.
- The additions support richer accent breakdowns aligned with Kachru’s Three Circles model.
- Private splits aim to minimize benchmark gaming by holding out evaluation data from public access.
Hugging Face’s Open ASR Leaderboard has added two new conversational speech datasets for Egyptian and Nigerian English, provided by Liva AI. The datasets total 1.34 hours for Egyptian English and 1.55 hours for Nigerian English, featuring spontaneous speech with punctuated, cased transcriptions that include disfluencies.
These additions follow earlier expansions that introduced private evaluation splits from Appen, DataOceanAI, and Voice Arena, including an Indian-accented English split. The new datasets continue the leaderboard’s effort to improve representation in ASR evaluations and reduce benchmark gaming by using held-out private data.
The Open ASR Leaderboard now groups accents using Kachru’s Three Circles framework, enabling comparisons between Inner Circle varieties (US, UK, Australia, Canada) and Outer/Expanding Circle varieties (India, Nigeria, Egypt). As more accents are added, the leaderboard plans to further split the Outer and Expanding Circles for finer-grained analysis.
Liva AI, which contributed the datasets, collects multimodal interaction data through consumer products, allowing it to capture natural, high-context behavior from real-world use rather than artificially elicited speech. This approach provides datasets that reflect authentic conversational patterns, including disfluencies and spontaneous speech.
- Nov 11, 2026 · Hugging Face
Hugging Face proposes token pooling-aware training to improve compressed multi-vector retrieval
Trust79 - Nov 11, 2026 · TechCrunch — AI
Fractile raises $720M Series C to develop AI chips and compiler stack
Trust79 - Nov 11, 2026 · The Verge — AI
Google’s Tabletop AI experiment lets users build custom collaborative apps via waitlist
Trust79