arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.13952cs.SDcs.AIeess.AS
收藏

基于大语言模型的强化学习音频视觉语音增强

LLM-Guided Reinforcement Learning for Audio-Visual Speech Enhancement

Department of Electrical Engineering, National Taiwan University, Taiwan(台湾大学电子工程系);Research Center for Information Technology Innovation, Academia Sinica, Taiwan(中央研究院资讯科技创新研究中心);Center for Hearing Research, University of California Irvine, USA(加州大学尔湾分校听力研究中心)

Chih-Ning Chen, Jen-Cheng Hou, Hsin-Min Wang, Shao-Yi Chien, Yu Tsao, Fan-Gang Zeng

AI总结:

本文提出基于大语言模型的可解释奖励模型的音频视觉语音增强框架,利用语言模型生成语音增强的自然语言描述,并通过情感分析模型转换为评分作为PPO奖励,提升语音质量评估效果。

摘要:

在现有的音频视觉语音增强(AVSE)方法中,如尺度不变信号噪声比(SI-SNR)和均方误差(MSE)等目标被广泛使用;然而,它们往往与感知质量相关性较差,并且对优化过程的可解释性有限。本文提出了一种基于强化学习的AVSE框架,结合大语言模型(LLM)的可解释奖励模型。音频LLM生成增强语音的自然语言描述,通过情感分析模型转换为1-5分的评分,作为PPO奖励对预训练的AVSE模型进行微调。与标量指标相比,LLM生成的反馈在语义上更丰富,并明确描述了语音质量的改进。在第4届COG-MHEAR AVSE挑战赛(AVSEC-4)数据集上的实验表明,所提出的方法在PESQ、STOI、神经质量指标和主观听测测试中均优于监督基线和DNSMOS基于的RL基线。

英文摘要:

In existing Audio-Visual Speech Enhancement (AVSE) methods, objectives such as Scale-Invariant Signal-to-Noise Ratio (SI-SNR) and Mean Squared Error (MSE) are widely used; however, their correlation with perceived speech quality is often suboptimal and provides limited interpretability for optimization. This work proposes a reinforcement learning-based AVSE framework with a Large Language Model (LLM)-based interpretable reward model. An audio LLM generates natural language descriptions of enhanced speech, which are converted by a sentiment analysis model into a 1-5 rating score serving as the PPO reward for fine-tuning a pretrained AVSE model. Compared with scalar metrics, LLM-generated feedback is semantically rich and explicitly describes speech quality improvements. Experiments on the AVSEC-4 dataset show that the proposed method outperforms a supervised baseline and a DNSMOS-based RL baseline in PESQ, STOI, neural quality metrics, and subjective listening tests.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑