Open ASR Leaderboard expands to Multilingual & Long-Form Tracks

AI Impact Summary

The Open ASR Leaderboard has expanded to include multilingual and long-form transcription tracks, reflecting a growing demand for ASR models beyond short-form English. Key findings highlight the dominance of Conformer encoder + LLM decoder models for accuracy, CTC/TDT decoders for speed, and the continued advantage of closed-source systems in long-form audio transcription due to domain tuning and optimization. This expansion represents a significant shift in the ASR landscape, demanding attention to model specialization and efficiency tradeoffs.