Egyptian and Nigerian English Accents Added to Open ASR Leaderboard

Hugging Face has announced a new collaboration with Liva AI that brings two additional private evaluation splits to the Open ASR Leaderboard. The datasets comprise 1.34 hours of spontaneous Egyptian‑accented English and 1.55 hours of spontaneous Nigerian‑accented English, both transcribed with punctuation, casing, and disfluency markers.

Richer accent breakdown for more trustworthy benchmarks

With these additions the leaderboard now reports performance across a broader set of accents, grouped using Kachru’s Three Circles model: Inner Circle (US, UK, Australia, Canada) versus Outer and Expanding Circles (India, Nigeria, Egypt). This finer granularity helps researchers spot systematic gaps that single‑accent averages hide.

Liva AI’s collection method — capturing multimodal interaction data from real consumer products — yields natural, high‑context speech that differs from scripted or elicited corpora. Because Liva controls both the environment and distribution, it can target specific model weaknesses across domains, languages, and interaction types.

Why it matters for GPU / AI infrastructure: More diverse, private test sets increase the compute demand for thorough model evaluation and for training models that generalize across under‑represented accents. Cloud GPU providers such as AiGpu can expect higher workload variability as teams run larger benchmark suites and fine‑tune models on these new accents.