Adding Benchmaxxer Repellant to the Open ASR Leaderboard

The Open ASR Leaderboard adds private datasets to prevent benchmaxxing and improve model evaluation robustness.

• •Appen Inc. and DataoceanAI provided English ASR datasets with scripted and conversational speech across multiple accents
• •Private datasets prevent benchmark-specific optimization and test-set contamination, addressing Goodhart's Law
• •Users can toggle public and private datasets and filter by speech style or accent
• •Metrics reveal performance gaps between controlled and real-world conversational speech
• •Models are evaluated on both public and private datasets after GitHub PR submission

This summary was automatically generated by AI based on the original article and may not be fully accurate.