Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Liu, Chang, Hu, Ya-Jun, Gao, Ying-Ying, Zhang, Shi-Lei, Ling, Zhen-Hua
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2509.18798
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908555200692224
author Liu, Chang
Hu, Ya-Jun
Gao, Ying-Ying
Zhang, Shi-Lei
Ling, Zhen-Hua
author_facet Liu, Chang
Hu, Ya-Jun
Gao, Ying-Ying
Zhang, Shi-Lei
Ling, Zhen-Hua
contents This paper proposes a GRPO-based approach to enhance the performance of large language model (LLM)-based text-to-speech (TTS) models by deriving rewards from an off-the-shelf automatic speech recognition (ASR) model. Compared to previous reinforcement learning methods for LLM-based TTS, our method requires no dedicated model for reward computation or training. Moreover, we design a composite reward function that combines character error rate (CER) with negative log-likelihood (NLL) obtained from the ASR model, providing more informative and accurate reward signals. We apply GRPO fine-tuning to pre-trained LLM-based TTS models and evaluate their zero-shot TTS performance. Experimental results show that the proposed method substantially improves both the intelligibility and naturalness of synthesized speech. Ablation studies and further analyses confirm the effectiveness of integrating the two reward components.
format Preprint
id arxiv_https___arxiv_org_abs_2509_18798
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Group Relative Policy Optimization for Text-to-Speech with Large Language Models
Liu, Chang
Hu, Ya-Jun
Gao, Ying-Ying
Zhang, Shi-Lei
Ling, Zhen-Hua
Audio and Speech Processing
This paper proposes a GRPO-based approach to enhance the performance of large language model (LLM)-based text-to-speech (TTS) models by deriving rewards from an off-the-shelf automatic speech recognition (ASR) model. Compared to previous reinforcement learning methods for LLM-based TTS, our method requires no dedicated model for reward computation or training. Moreover, we design a composite reward function that combines character error rate (CER) with negative log-likelihood (NLL) obtained from the ASR model, providing more informative and accurate reward signals. We apply GRPO fine-tuning to pre-trained LLM-based TTS models and evaluate their zero-shot TTS performance. Experimental results show that the proposed method substantially improves both the intelligibility and naturalness of synthesized speech. Ablation studies and further analyses confirm the effectiveness of integrating the two reward components.
title Group Relative Policy Optimization for Text-to-Speech with Large Language Models
topic Audio and Speech Processing
url https://arxiv.org/abs/2509.18798