반응형
SFM-TTS: Lightweight and Rapid Speech Synthesis with Flexible Shortcut Flow MatchingFlow matching Text-to-Speech model은 small step에서 generation quality가 떨어짐SFM-TTSStandard Gaussian distribution을 linear interpolation을 통해 ground-truth distribution으로 transform추가적으로 Fast Linear Attention을 활용해 parameter 수를 절감논문 (ICASSP 2026) : Paper Link1. IntroductionFlow Matching은 noise에서 data로의 continuous-time tra..
Paper/TTS
2026. 5. 8. 10:47
반응형
