1、OutlineOutline The algorithm of FastSpeech By Xu Tan, Microsoft Research Asia The optimization of FastSpeech By Dabi Ahn, NVIDIA About text to speech systemAbout text to speech system TTS Frontend Acoustic Model VocoderTextSpeech FastSpeechF s t s p iy ch Phoneme Mel-spectrogram About About FastSpee
2、chFastSpeech A fast, robust, controllable, high-quality and end-to-end text to speech (TTS) system FastSpeech: Fast, Robust and Controllable Text to Speech, NeurIPS 2019 1 FastSpeech 2: Fast and High-Quality End-to-End Text to Speech, ICLR 2021 submission 2 Widely supported by the community, and dep
3、loyed in Microsoft Azure TTS service to support all the languages 1 https:/proceedings.neurips.cc/paper/2019/file/f63f65b503e22cb970527f23c9ad7db1-Paper.pdf 2 https:/ BackgroundBackground End-to-end neural TTS such as Tacotron 2 1, Deep Voice 3 2, and Transformer TTS 3 has achieved good voice quality. However, end-to-end neural TTS still suffers from the following problems Slow inference speed: au