ProfASR-Bench is a professional-talk evaluation suite for context-conditioned ASR in high-stakes applications across Finance, Medicine, Legal, and Technology. Each example pairs a natural-language prompt (domain cue and/or speaker profile) with an entity-rich target utterance, enabling controlled measurement of context-conditioned recognition.
Key Finding - The Context-Utilization Gap (CUG): Current systems are nominally… See the full description on the dataset page: https://huggingface.co/datasets/prdeepakbabu/ProfASR-Bench.
