A two-stage post-training pipeline of SFT followed by editing-oriented GRPO on unpaired data improves speech editing consistency and zero-shot TTS quality.
Edit Content, Preserve Acoustics
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
CosyEdit2: Speech-Editing-Oriented Reinforcement Learning Unlocks Better Zero-Shot TTS
A two-stage post-training pipeline of SFT followed by editing-oriented GRPO on unpaired data improves speech editing consistency and zero-shot TTS quality.