PapersBenchmarks
Jiutian Artificial Intelligence Research Institute
23 Sep 2025
audio-and-speech-processingelectrical-engineering

Group Relative Policy Optimization for Text-to-Speech with Large Language Models

University of Science and Technology of China logoUniversity of Science and Technology of ChinaChina MobileiFlytek ResearchNational Engineering Research Center of Speech and Language Information ProcessingJiutian Artificial Intelligence Research Institute
This research introduces Group Relative Policy Optimization (GRPO) with a novel off-the-shelf ASR-derived composite reward to fine-tune large language model (LLM)-based Text-to-Speech (TTS) models. The approach enhances speech intelligibility and naturalness, achieving consistent improvements in character/word error rates and Mean Opinion Scores across multiple languages and diverse LLM-TTS architectures.
There are no more papers matching your filters at the moment.