SoundMind: RL-Incentivized Logic Reasoning for Audio-Language Models

2025-06-15

AI Summary

Current large audio-language models (ALMs) exhibit severe deficiencies in audio-modal logical reasoning. To address this, we introduce ALR—the first benchmark dataset specifically designed for complex audio logical reasoning, comprising 6,446 annotated samples. We further propose SoundMind, a rule-guided reinforcement learning framework that enables interpretable and controllable logical capability injection into ALMs for the first time. Our approach integrates rule-constrained RL training, audio-text multimodal alignment modeling, and fine-tuning of the Qwen2.5-Omni-7B architecture. On the ALR benchmark, SoundMind achieves state-of-the-art performance, substantially enhancing ALMs’ deep semantic and logical understanding of audio inputs. Both the ALR dataset and SoundMind implementation are fully open-sourced, establishing a foundational resource and methodological paradigm for auditory intelligence research.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Knowledge Representation and Reasoning: Common-Sense Reasoning

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

While large language models have shown reasoning capabilities, their application to the audio modality, particularly in large audio-language models (ALMs), remains significantly underdeveloped. Addressing this gap requires a systematic approach, involving a capable base model, high-quality reasoning-oriented audio data, and effective training algorithms. In this study, we present a comprehensive solution: we introduce the Audio Logical Reasoning (ALR) dataset, consisting of 6,446 text-audio annotated samples specifically designed for complex reasoning tasks. Building on this resource, we propose SoundMind, a rule-based reinforcement learning (RL) algorithm tailored to endow ALMs with deep bimodal reasoning abilities. By training Qwen2.5-Omni-7B on the ALR dataset using SoundMind, our approach achieves state-of-the-art performance in audio logical reasoning. This work highlights the impact of combining high-quality, reasoning-focused datasets with specialized RL techniques, advancing the frontier of auditory intelligence in language models. Our code and the proposed dataset are available at https://github.com/xid32/SoundMind.

Problem

• Enhancing audio-language models' reasoning in audio modality
• Developing high-quality reasoning-focused audio-text datasets
• Applying RL to improve bimodal reasoning in ALMs

Innovation

• Introduces Audio Logical Reasoning dataset
• Proposes rule-based RL algorithm SoundMind
• Trains Qwen2.5-Omni-7B for audio reasoning