Experiments with silenced and replacement audio suggest that RT-OPD strengthens the student's reliance on acoustic evidence, and the 3B model achieves 72.72% accuracy on MMAU, the highest among the compared 3B models and competitive with several 7B and 8B models.
Abstract
Audio-language models (ALMs) can exploit textual shortcuts to answer questions while overlooking acoustic evidence, weakening audio understanding. On-policy distillation (OPD) trains compact ALMs by supervising student-generated responses with teacher predictions, but does not explicitly distinguish acoustic support from linguistic predictability. We propose Reward-Tilted On-Policy Distillation (RT-OPD) to strengthen acoustic grounding. Given the same question and student-generated text, a frozen teacher predicts the next token with and without audio inputs. Their log-probability contrast defines a reward that reshapes the teacher distribution for reverse-KL distillation, emphasizing the additional evidence provided by audio. Across two compact students and three benchmarks, RT-OPD consistently outperforms Vanilla OPD. Experiments with silenced and replacement audio further suggest that RT-OPD strengthens the student's reliance on acoustic evidence. Our 3B model achieves 72.72% accuracy on MMAU, the highest among the compared 3B models and competitive with several 7B and 8B models. Code and model checkpoints are available at https://github.com/KaiyangLi1992/RT-OPD.
Omni-modal large language models deployed in real-world environments encounter external noise that can interfere with their perception and understanding of multimodal inputs. We study their robustness in audio-visual understanding, focusing on question answering under environmental noise and competing speech. The chall...
Xing-Ming Shui, Da-Peng Chen, Bo-Wei Liu et al.· arXiv.org· 0 citations
It is suggested that auditory reasoning can benefit from training strategies tailored to audio signals and objective answer verification, and they provide empirical evidence for applying RLVR to large audio–language models.
TTRSD separates update direction from update position, determined by group-relative advantages, from update position, determined by visual sensitivity, without requiring ground-truth labels, external verifiers, or a separate teacher, demonstrating cross-dataset generalization while preserving inherent reasoning integri...
Joint-Output On-Policy Distillation (JO-OPD) is proposed, which distills the model's stronger S2T policy into joint generation using student-generated S2TS trajectories, which reduces OMG across two interleaved generation architectures.
Da-Xin Tan, De-Hua Tao, Chen-Xi Deng et al.· 0 citations
Large audio language models answer questions about speech, sound, and music, yet their accuracy drops sharply on tasks that need fine-grained acoustic analysis. Judging which of two speakers has the higher pitch demands iterative signal-level reasoning that a content question does not. Current models spend the same com...
Long multilingual conversational spoken question answering requires systems to balance long-range transcript semantics with sparse acoustic and speaker-sensitive cues. We present the Bairong system for the MLC-SLM 2026 Challenge, where a diarization-ASR front-end produces speaker-attributed transcripts and a dynamic ev...
Shang-Kun Huang, Jun-Chao Hu, Hua Shen et al.· 2nd Workshop on Multilingual...· 1 citation