Are Audio-Language Models Listening? Audio-Specialist Heads for Adaptive Audio Steering

TL;DR

This paper identifies and enhances audio-specific attention mechanisms in large audio-language models to improve their ability to utilize audio evidence effectively, leading to better performance without retraining.

Contribution

It introduces a mechanistic interpretability approach to locate audio-specialist attention heads and applies an inference-time intervention to boost audio engagement.

Findings

Increased audio attention correlates with model output changes.

Intervention improves accuracy by up to +8.0 percentage points.

Method works on Qwen-based LALMs without retraining.

Abstract

Multimodal large language models can exhibit text dominance, over-relying on linguistic priors instead of grounding predictions in non-text inputs. One example is large audio-language models (LALMs) where decisive audio evidence can be under-utilized even when it contains important information. To address this issue we use mechanistic interpretability to identify a small set of audio-specialist attention heads whose audio attention yields a ``listening'' signal. We show that this signal increases when audio evidence affects the model's output, providing an indicator of audio engagement under standard prompting. Leveraging this localization, we construct an audio--silence steering direction and apply an inference-time activation intervention to the final representation, amplifying the model's audio effect. To demonstrate the utility of this intervention, we show on MMAU that this improves accuracy by up to +8.0 percentage points on two Qwen-based LALMs, without any parameter updates.