Saved in:
Bibliographic Details
Main Authors: Zhang, Haoyu, Guo, Jiaxian, Iwasawa, Yusuke, Matsuo, Yutaka
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.05478
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911391090212864
author Zhang, Haoyu
Guo, Jiaxian
Iwasawa, Yusuke
Matsuo, Yutaka
author_facet Zhang, Haoyu
Guo, Jiaxian
Iwasawa, Yusuke
Matsuo, Yutaka
contents Large Audio Language Models (LALMs) demonstrate impressive general audio understanding, but once deployed, they are static and fail to improve with new real-world audio data. As traditional supervised fine-tuning is costly, we introduce a novel framework for test-time audio understanding, AQA-TTRL, where an LALM evolves on-the-fly using only unlabeled test data. It first generates pseudo-labels from the prediction via majority voting, then optimizes the model via reinforcement learning. To handle the inherent noise in these self-generated labels, we introduce a confidence-based weighting method to adjust training signals. Furthermore, a multiple-attempt sampling operation mitigates advantage collapse and stabilizes training. On the MMAU (test-mini/test), MMAR, and MMSU benchmarks, AQA-TTRL achieves significant average improvements of 4.42% for the Qwen2.5-Omni 7B model and 11.04% for the 3B model. Notably, the adapted 3B model consistently outperforms the direct inference of the unadapted 7B model, highlighting the effectiveness of previously unexplored test-time adaptations in audio understanding.
format Preprint
id arxiv_https___arxiv_org_abs_2510_05478
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AQA-TTRL: Self-Adaptation in Audio Question Answering with Test-Time Reinforcement Learning
Zhang, Haoyu
Guo, Jiaxian
Iwasawa, Yusuke
Matsuo, Yutaka
Audio and Speech Processing
Large Audio Language Models (LALMs) demonstrate impressive general audio understanding, but once deployed, they are static and fail to improve with new real-world audio data. As traditional supervised fine-tuning is costly, we introduce a novel framework for test-time audio understanding, AQA-TTRL, where an LALM evolves on-the-fly using only unlabeled test data. It first generates pseudo-labels from the prediction via majority voting, then optimizes the model via reinforcement learning. To handle the inherent noise in these self-generated labels, we introduce a confidence-based weighting method to adjust training signals. Furthermore, a multiple-attempt sampling operation mitigates advantage collapse and stabilizes training. On the MMAU (test-mini/test), MMAR, and MMSU benchmarks, AQA-TTRL achieves significant average improvements of 4.42% for the Qwen2.5-Omni 7B model and 11.04% for the 3B model. Notably, the adapted 3B model consistently outperforms the direct inference of the unadapted 7B model, highlighting the effectiveness of previously unexplored test-time adaptations in audio understanding.
title AQA-TTRL: Self-Adaptation in Audio Question Answering with Test-Time Reinforcement Learning
topic Audio and Speech Processing
url https://arxiv.org/abs/2510.05478