Measuring Audio's Impact on Correctness: Audio-Contribution-Aware Post-Training of Large Audio Language Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: He, Haolin, Du, Xingjian, Sun, Renhe, Dai, Zheqi, Xiao, Yujia, Yang, Mingru, Zhou, Jiayi, Li, Xiquan, Liu, Zhengxi, Liang, Zining, Wu, Chunyat, He, Qianhua, Lee, Tan, Chen, Xie, Zheng, Wei-Long, Wang, Weiqiang, Plumbley, Mark, Liu, Jian, Kong, Qiuqiang
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866917323849334784
author He, Haolin
Du, Xingjian
Sun, Renhe
Dai, Zheqi
Xiao, Yujia
Yang, Mingru
Zhou, Jiayi
Li, Xiquan
Liu, Zhengxi
Liang, Zining
Wu, Chunyat
He, Qianhua
Lee, Tan
Chen, Xie
Zheng, Wei-Long
Wang, Weiqiang
Plumbley, Mark
Liu, Jian
Kong, Qiuqiang
author_facet He, Haolin
Du, Xingjian
Sun, Renhe
Dai, Zheqi
Xiao, Yujia
Yang, Mingru
Zhou, Jiayi
Li, Xiquan
Liu, Zhengxi
Liang, Zining
Wu, Chunyat
He, Qianhua
Lee, Tan
Chen, Xie
Zheng, Wei-Long
Wang, Weiqiang
Plumbley, Mark
Liu, Jian
Kong, Qiuqiang
contents Large Audio Language Models (LALMs) represent an important frontier in multimodal AI, addressing diverse audio tasks. Recently, post-training of LALMs has received increasing attention due to significant performance improvements over foundation models. While single-stage post-training such as reinforcement learning (RL) has demonstrated promising results, multi-stage approaches such as supervised fine-tuning (SFT) followed by RL remain suboptimal. The allocation of data across multiple training stages to maximize LALM capabilities has not been fully explored, and large-scale, high-quality datasets for such research are also lacking. To address these problems, we firstly present AudioMCQ, a comprehensive audio multiple-choice question dataset comprising 571k samples with two kinds of chain-of-thought annotations. Secondly, we investigate the prevalent zero audio-contribution phenomenon in LALMs, where models derive correct answers solely from textual information without processing audio content. We propose Audio-Contribution Filtering to partition data into weak and strong audio-contribution subsets. Based on these insights, we develop two effective post-training paradigms: Weak-to-Strong (SFT on weak audio-contribution data followed by RL on strong audio-contribution data) and Mixed-to-Strong (SFT on mixed audio-contribution data followed by RL on strong audio-contribution data). We achieve first place in the DCASE 2025 Audio-Question-Answering challenge by using AudioMCQ. Additionally, leveraging our dataset with different training strategies, we achieve 78.2\% on MMAU-test-mini, 75.6\% on MMAU, 67.1\% on MMAR, and 70.7\% on MMSU, establishing new state-of-the-art performance.
format Preprint
id arxiv_https___arxiv_org_abs_2509_21060
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Measuring Audio's Impact on Correctness: Audio-Contribution-Aware Post-Training of Large Audio Language Models
He, Haolin
Du, Xingjian
Sun, Renhe
Dai, Zheqi
Xiao, Yujia
Yang, Mingru
Zhou, Jiayi
Li, Xiquan
Liu, Zhengxi
Liang, Zining
Wu, Chunyat
He, Qianhua
Lee, Tan
Chen, Xie
Zheng, Wei-Long
Wang, Weiqiang
Plumbley, Mark
Liu, Jian
Kong, Qiuqiang
Audio and Speech Processing
Large Audio Language Models (LALMs) represent an important frontier in multimodal AI, addressing diverse audio tasks. Recently, post-training of LALMs has received increasing attention due to significant performance improvements over foundation models. While single-stage post-training such as reinforcement learning (RL) has demonstrated promising results, multi-stage approaches such as supervised fine-tuning (SFT) followed by RL remain suboptimal. The allocation of data across multiple training stages to maximize LALM capabilities has not been fully explored, and large-scale, high-quality datasets for such research are also lacking. To address these problems, we firstly present AudioMCQ, a comprehensive audio multiple-choice question dataset comprising 571k samples with two kinds of chain-of-thought annotations. Secondly, we investigate the prevalent zero audio-contribution phenomenon in LALMs, where models derive correct answers solely from textual information without processing audio content. We propose Audio-Contribution Filtering to partition data into weak and strong audio-contribution subsets. Based on these insights, we develop two effective post-training paradigms: Weak-to-Strong (SFT on weak audio-contribution data followed by RL on strong audio-contribution data) and Mixed-to-Strong (SFT on mixed audio-contribution data followed by RL on strong audio-contribution data). We achieve first place in the DCASE 2025 Audio-Question-Answering challenge by using AudioMCQ. Additionally, leveraging our dataset with different training strategies, we achieve 78.2\% on MMAU-test-mini, 75.6\% on MMAU, 67.1\% on MMAR, and 70.7\% on MMSU, establishing new state-of-the-art performance.
title Measuring Audio's Impact on Correctness: Audio-Contribution-Aware Post-Training of Large Audio Language Models
topic Audio and Speech Processing
url https://arxiv.org/abs/2509.21060