Tag
The paper introduces Hear2Act, a unified benchmark for evaluating how prosodic cues affect task-oriented dialogue decisions, finding that prosody matters when lexical evidence is insufficient and that audio LLMs benefit from explicit concern representation for actions.