Tongyi Releases FlowTTS-GRPO Voice Cloning Model

udmrzn · x · 2026-07-14

Tongyi Voice AI introduced its FlowTTS-GRPO voice cloning model. Traditional voice cloning models often struggle to balance text accuracy, target timbre fidelity, and audio clarity/stability.

By applying Online Reinforcement Learning (Online RL) to flow-matching TTS, this new model successfully achieves simultaneous optimization across all three of these core dimensions.

Original post →

More from Research

Research channel →