반응형
[Paper 리뷰] NCF-TTS: Enhancing Flow Matching based Text-to-Speech with Neighborhood Consistency Flow
NCF-TTS: Enhancing Flow Matching based Text-to-Speech with Neighborhood Consistency FlowDiffusion-based Text-to-Speech는 추론 속도의 한계와 guidance method에 대한 incompatibility가 존재함NCF-TTSLarge-step sampling을 stablize 하는 Neighborhood Consistency Flow를 활용Conditional, unconditional supervision을 training process로 unify 하는 embedded guidance objective를 도입하고 flow matching supervision과 NCF consistency loss를 join..
Paper/TTS
2026. 5. 6. 11:04
반응형