When Scaling Fails: Mitigating Audio Perception Decay of LALMs via Multi-Step Perception-Aware Reasoning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Mao, Ruixiang, Ma, Xiangnan, Chen, Dan, Zhu, Ziming, Ge, Yuan, Hao, Aokai, Zhao, Haishu, Huo, Yifu, Yang, Qing, Chang, Kaiyan, Liu, Xiaoqian, Wang, Chenglong, He, Qiaozhi, Xiao, Tong, Zhu, Jingbo
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908861160488960
author Mao, Ruixiang
Ma, Xiangnan
Chen, Dan
Zhu, Ziming
Ge, Yuan
Hao, Aokai
Zhao, Haishu
Huo, Yifu
Yang, Qing
Chang, Kaiyan
Liu, Xiaoqian
Wang, Chenglong
He, Qiaozhi
Xiao, Tong
Zhu, Jingbo
author_facet Mao, Ruixiang
Ma, Xiangnan
Chen, Dan
Zhu, Ziming
Ge, Yuan
Hao, Aokai
Zhao, Haishu
Huo, Yifu
Yang, Qing
Chang, Kaiyan
Liu, Xiaoqian
Wang, Chenglong
He, Qiaozhi
Xiao, Tong
Zhu, Jingbo
contents Test-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 post-training for direct answering. To investigate it, we introduce CAFE, an evaluation framework designed to precisely quantify audio reasoning errors. Evaluation results reveal LALMs struggle with perception during reasoning and encounter a critical bottleneck: reasoning performance suffers from audio perception decay as reasoning length extends. To address it, we propose MPAR$^2$, a paradigm that encourages dynamic perceptual reasoning and decomposes complex questions into perception-rich sub-problems. Leveraging reinforcement learning, MPAR$^2$ improves perception performance on CAFE from 31.74% to 63.51% and effectively mitigates perception decay, concurrently enhancing reasoning capabilities to achieve a significant 74.59% accuracy on the MMAU benchmark. Further analysis demonstrates that MPAR$^2$ reinforces LALMs to attend to audio input and dynamically adapts reasoning budget to match task complexity.
format Preprint
id arxiv_https___arxiv_org_abs_2603_02266
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle When Scaling Fails: Mitigating Audio Perception Decay of LALMs via Multi-Step Perception-Aware Reasoning
Mao, Ruixiang
Ma, Xiangnan
Chen, Dan
Zhu, Ziming
Ge, Yuan
Hao, Aokai
Zhao, Haishu
Huo, Yifu
Yang, Qing
Chang, Kaiyan
Liu, Xiaoqian
Wang, Chenglong
He, Qiaozhi
Xiao, Tong
Zhu, Jingbo
Sound
Artificial Intelligence
Audio and Speech Processing
Test-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 post-training for direct answering. To investigate it, we introduce CAFE, an evaluation framework designed to precisely quantify audio reasoning errors. Evaluation results reveal LALMs struggle with perception during reasoning and encounter a critical bottleneck: reasoning performance suffers from audio perception decay as reasoning length extends. To address it, we propose MPAR$^2$, a paradigm that encourages dynamic perceptual reasoning and decomposes complex questions into perception-rich sub-problems. Leveraging reinforcement learning, MPAR$^2$ improves perception performance on CAFE from 31.74% to 63.51% and effectively mitigates perception decay, concurrently enhancing reasoning capabilities to achieve a significant 74.59% accuracy on the MMAU benchmark. Further analysis demonstrates that MPAR$^2$ reinforces LALMs to attend to audio input and dynamically adapts reasoning budget to match task complexity.
title When Scaling Fails: Mitigating Audio Perception Decay of LALMs via Multi-Step Perception-Aware Reasoning
topic Sound
Artificial Intelligence
Audio and Speech Processing
url https://arxiv.org/abs/2603.02266