When Scaling Fails: Mitigating Audio Perception Decay of LALMs via Multi-Step Perception-Aware Reasoning
Abstract
AbstractTest-Time Scaling has shown notable efficacy in addressing complex problems through scaling inference compute. However, within Large Audio-Language Models (LALMs), an unintuitive phenomenon exists: post-training models for structured reasoning trajectories results in marginal or even negative gains compared to training for direct answering. To investigate it, we introduce CAFE, an evaluation framework designed to quantify audio reasoning errors. Evaluation results reveal that LALMs encounter a critical bottleneck: Audio Perception Decay, which compromises reasoning performance. To address it, we propose MPAR2, a strategy that encourages a step-by-step reasoning process with rich perceptual awareness. With reinforcement learning, MPAR2 improves performance on CAFE from 31.74% to 63.51% and effectively mitigates audio perception decay, concurrently enhancing reasoning ability to achieve a strong 74.59% accuracy on the MMAU benchmark. Further analysis shows that MPAR2 optimizes audio attention and enhances reasoning token budget efficiency.