Open ASR Leaderboard adds private Egyptian and Nigerian English test sets from Liva AI

The Open ASR Leaderboard has added two new private evaluation splits from Liva AI covering Egyptian and Nigerian accented English, according to a Hugging Face blog post published November 6, 2026. The post is authored by Eric Bezzam, Ashley Mo and Aoi Otani.

The new data

Both splits are conversational, spontaneous speech. Transcriptions are punctuated, cased, and include disfluencies.

Dataset	Accent	Duration (hours)	Style
Liva AI Conversational EG	Egyptian	1.34	Conversational, spontaneous
Liva AI Conversational NG	Nigerian	1.55	Conversational, spontaneous

The maintainers say the addition is part of an effort to improve representation in ASR evaluations (more accents and more languages) and to improve trust, using new private data to minimize "benchmaxxing" and adding new data providers.

Where this fits in the private data program

The post recounts the leaderboard's earlier private data additions. A few months ago, the team added 11 splits from Appen and DataOceanAI, covering scripted and conversational speech from five accents. It later added one private split for Indian-accented English with Voice Arena. The Liva AI splits are the newest addition.

New accent breakdown in the Private tab

With Egyptian and Nigerian English included, the leaderboard now presents accent results based on Kachru's Three Circles. The post credits Hume AI for the idea, from its own leaderboard. The Open ASR Leaderboard will report two groupings:

• US vs. non-US: non-US covers UK, Australia, Canada, India, Nigeria and Egypt.
• Inner Circle vs. Outer & Expanding Circles: Inner Circle covers US, UK, Australia and Canada. Outer & Expanding Circles cover India, Nigeria and Egypt.

The maintainers say they will split the Outer and Expanding Circles as more accents join.

About Liva AI

Liva AI collects multimodal interaction data through consumer products used by real people. According to the post, because Liva owns both the collection environment and distribution, it can design products around specific model gaps across domains, industries, languages, modalities and interaction types. The post says this gives access to natural, high-context behavior from real product use, rather than artificially elicited or task-based interactions.

What this means

The Private tab now includes Egyptian and Nigerian accents alongside the existing private splits, so the non-US grouping spans six accents: UK, Australia, Canada, India, Nigeria and Egypt. Private data is the maintainers' stated mechanism for limiting benchmaxxing, since the test material is not public. The Three Circles grouping gives a standardized way to compare Inner Circle accents against Outer and Expanding Circle accents, and the maintainers plan to refine it as more accents are added. The new Liva AI splits are conversational and include disfluencies, which makes them a test of spontaneous speech rather than scripted reading.