Open ASR Leaderboard Adds Egyptian and Nigerian English Accents

New Accents Added to Open ASR Leaderboard

The Open ASR Leaderboard has announced the inclusion of two new English accents: Egyptian and Nigerian. This expansion is a result of a new collaboration with Liva AI, which provided the private data for these accents. The addition follows previous expansions that included private data for Indian-accented English and other accents from Appen and DataOceanAI.

Improving ASR Evaluation Representation

This update is part of an ongoing effort to enhance representation in Automatic Speech Recognition (ASR) evaluations. The goal is to include more diverse accents and languages, as well as to improve the trustworthiness of evaluations by adding new private data to minimize 'benchmaxxing'. The initiative also involves incorporating new data providers to broaden the scope of evaluation data.

Richer Accent Breakdown

With the integration of Egyptian and Nigerian English, the Open ASR Leaderboard can now offer a more detailed breakdown of accents. This breakdown is structured based on Kachru’s Three Circles model, a framework for classifying the spread of English. Future plans include further splitting the Outer and Expanding Circles as more accents are added to the leaderboard.

Liva AI's Data Collection Approach

Liva AI collects multimodal interaction data from consumer products used by real individuals. By controlling both the collection environment and distribution, Liva AI can design products that address specific model gaps across various domains, industries, languages, modalities, and interaction types. This method provides access to natural, high-context behavioral data, as opposed to artificially elicited or task-based interactions.