Beyond Transcripts: Iterative Peer-Editing with Audio Unlocks High-Quality Human Summaries of Conversational Speech

In three linesComparative study of 10 annotation workflows for conversational speech summarization. Audio-based summaries are less informative than transcript-based ones, but iterative peer-editing with audio mitigates this gap. Validates this approach for creating benchmarks incorporating lexical and prosodic information.

Summary generated by Claude — human-verified