EvA: An Evidence-First Audio Understanding Paradigm for LALMs

TL;DR

EvA introduces a dual-path architecture that enhances audio evidence preservation in LALMs, significantly improving perception accuracy in complex acoustic scenes by combining Whisper and CED-Base.

Contribution

The paper proposes EvA, an evidence-first paradigm with a novel fusion method and a large-scale dataset, advancing audio understanding by emphasizing evidence preservation.

Findings

EvA achieves state-of-the-art open-source perception scores on multiple benchmarks.

EvA outperforms Kimi-Audio-7B across all metrics, especially in perception-heavy splits.

Preserving acoustic evidence before reasoning improves LALMs' performance in complex scenes.

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 call this failure the evidence bottleneck: state-of-the-art systems show larger deficits in evidence extraction than in downstream reasoning, suggesting that the main limitation lies in upstream perception rather than reasoning policy. To address this problem, we propose EvA (Evidence-First Audio), a dual-path architecture that combines Whisper and CED-Base through non-compressive, time-aligned fusion. EvA first aggregates intermediate CED layers to preserve multi-scale acoustic cues, then aligns the aggregated CED features to the Whisper timeline and adds the two streams without changing sequence length. We also build EvA-Perception, a large-scale open-source training set with about 54K event-ordered captions (150 h) and about 500K QA pairs. Under a unified zero-shot protocol, EvA achieves the best open-source Perception scores on MMAU, MMAR, and MMSU, and improves over Kimi-Audio-7B on all reported metrics, with the largest gains on perception-heavy splits. These results support the evidence-first hypothesis: stronger audio understanding depends on preserving acoustic evidence before reasoning.