Tag
The paper proposes DiaWhisper-DPO, an end-to-end model for transcription and role attribution in clinical interviews using failure-mined preference optimization, achieving high accuracy and reducing errors compared to cascaded baselines.