arxivcs.LG2026-07-23
X$^3$-OPD: Distilling Reasoning into Large Audio-Language Models via On-Policy Alignment
Dongjie Fu, Di Cao, Xize Cheng, Zihan Zhang, Wenxu Jia, Yifu Chen, et al.
While large audio-language models have achieved remarkable progress in auditory perception, they still lag behind text-based large language models in deep logical reasoning, primarily due to the scarcity of high-quality audio reasoning data. To bridge this gap, we propose X$^3$-O…