반응형
[Paper 리뷰] SANE-TTS: Stable and Natural End-to-End Multilingual Text-to-Speech
SANE-TTS: Stable and Natural End-to-End Multilingual Text-to-SpeechStable 하고 natural 한 end-to-end multilingual text-to-speech 모델이 필요함SANE-TTSMultilingual synthesis의 naturalness를 향상하기 위해 domain adversarial training을 도입추가적으로 speaker regularization loss를 적용하여 duration predictor의 speaker embedding을 zero-vector로 대체해 cross-lingual synthesis를 stablize 함논문 (INTERSPEECH 2021) : Paper Link1. IntroductionM..
Paper/TTS
2024. 6. 4. 10:24
반응형