@inproceedings{mao-etal-2026-scaling, title = "When Scaling Fails: Mitigating Audio Perception Decay of {LALM}s via Multi-Step Perception-Aware Reasoning", author = "Mao, Ruixiang and Ma, Xiangnan and Chen, Dan and Zhu, Ziming and Ge, Yuan and Zhao, Haishu and Hao, AoKai and Huo, Yifu and Yang, Qing and Chang, Kaiyan and Liu, Xiaoqian and Wang, Chenglong and Shan, Weiqiao and He, Qiaozhi and Li, Changliang and Yu, Zhengtao and Xiao, Tong and Zhu, JingBo", editor = "Lee, Hung-Yi and Dev, Sunipa and Li, Jessy and Elliott, Desmond", booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing", month = oct, year = "2026", address = "Budapest, Hungary", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2026.emnlp-main.495/", pages = "11370--11393", ISBN = "979-8-89176-459-0" }