xg-chu / fs2_tts

TTS system base on FastSpeech2 and MelGAN.

FS2_TTS

TTS system base on FastSpeech2 and MelGAN.

Build

1. Requirements:
   • python 3.8.5, flask 1.1.2, tensorflow 2.3.1, react v17.0.1, node v14.15.1, pypinyin 0.40.0, soundfile 0.10.3.post1, scipy 1.5.4
2. Environments:
   • Build virtual environments and install flask.

   mkdir CN_FS2_TTS
   cd CN_FS2_TTS
   python3 -m venv flask_venv
   . flask_venv/bin/activate
   pip install Flask
   

   • Build React environments.

   npx create-react-app react_front_end
   cd react_front_end/src
   rm -f *
   cd ../
   cp ../react_front_end_src/* src/
   npm install @material-ui/core
   npm install --save react-router-dom
   

   and add the following codes to "scripts" in 'package.json' in 'react_front_end'.

   "prebuild": "rm ../templates/index.html && rm -rf ../static",
   "postbuild": "mv build/index.html ../templates/ && mv build/static ../static && mkdir ../static/tf_outputs",
   

fs2_tts's Issues

当文本长度较长时，后面的语音合成不出来。

测试了长度大概在42以后就无法识别了，请教是有参数控制吗？或者是fs2模型不支持长文本呢