Investigating Modality Contribution in Audio LLMs for Music

2025-09-24

AI Summary

This study investigates whether audio large language models (Audio LLMs) genuinely leverage the audio modality—rather than relying solely on textual reasoning—in music dialogues. To this end, we extend the Shapley-value-based MM-SHAP framework to audio–text multimodal settings for the first time and apply it with the MuChoMusic benchmark to quantify each modality’s attribution to model outputs. Results reveal that while overall audio contribution is low—and high-accuracy models predominantly depend on text—the audio modality is consistently utilized for precise localization of critical acoustic events (e.g., onsets, rhythmic transitions), confirming its functional role. This work introduces the first modality-attribution interpretability method tailored to Audio LLMs, exposing an implicit text bias in current evaluation protocols and providing both theoretical tools and empirical evidence for building trustworthy multimodal music understanding systems.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Audio Large Language Models (Audio LLMs) enable human-like conversation about music, yet it is unclear if they are truly listening to the audio or just using textual reasoning, as recent benchmarks suggest. This paper investigates this issue by quantifying the contribution of each modality to a model's output. We adapt the MM-SHAP framework, a performance-agnostic score based on Shapley values that quantifies the relative contribution of each modality to a model's prediction. We evaluate two models on the MuChoMusic benchmark and find that the model with higher accuracy relies more on text to answer questions, but further inspection shows that even if the overall audio contribution is low, models can successfully localize key sound events, suggesting that audio is not entirely ignored. Our study is the first application of MM-SHAP to Audio LLMs and we hope it will serve as a foundational step for future research in explainable AI and audio.

Problem

• Investigating whether Audio LLMs truly listen to audio or rely on text
• Quantifying modality contribution to model predictions using MM-SHAP framework
• Evaluating audio localization capability despite low overall audio contribution

Innovation

• Adapted MM-SHAP framework for modality contribution
• Evaluated two Audio LLMs on the MuChoMusic benchmark
• Quantified audio and text modality contributions using Shapley values