EvA: An Evidence-First Audio Understanding Paradigm for LALMs 事件

EvA: An Evidence-First Audio Understanding Paradigm for LALMs arXiv:2603.27667v2 Announce Type: replace-cross Abstract: Large Audio Language Models (LALMs) still struggle in complex acoustic scenes because they often fail to preserve task-relevant acoustic evidence before reasoning begins. We identify this error pattern as the evidence bottleneck: state-of-the-art systems show larger deficits in acoustic evidence extraction than in downstream reasoning, suggesting that upstream perception is of