Add TTS with VITS

[csukuangfj]

No description provided.

[csukuangfj]

Usage of this PR

Build sherpa-onnx with this PR

mkdir build
cd build
cmake ..
make -j 

(py38) fangjuns-MacBook-Pro:build fangjun$ ls -lh bin/sherpa-onnx-offline-tts
-rwxr-xr-x  1 fangjun  staff    36K Oct 13 19:17 bin/sherpa-onnx-offline-tts

Download the model files

wget https://huggingface.co/csukuangfj/vits-ljs/resolve/main/vits-ljs.onnx
wget https://huggingface.co/csukuangfj/vits-ljs/resolve/main/lexicon.txt
wget https://huggingface.co/csukuangfj/vits-ljs/resolve/main/tokens.txt

Run the model

./bin/sherpa-onnx-offline-tts \
  --vits-model=./vits-ljs.onnx \
  --vits-lexicon=./lexicon.txt \
  --vits-tokens=./tokens.txt \
  'Success is not final, failure is not fatal, it is the courage to continue that counts!'

It will generate a file

$ ls -lh t.pcm
-rw-r--r--  1 fangjun  staff   489K Oct 13 19:16 t.pcm

Please use the following command to convert it to a wave file:

 sox -t raw -r 22050 -b 32 -e floating-point -c 1 ./t.pcm ./1.wav

And you will see

$ ls -lh 1.wav
-rw-r--r--  1 fangjun  staff   489K Oct 13 19:16 1.wav

$ soxi 1.wav

Input File     : '1.wav'
Channels       : 1
Sample Rate    : 22050
Precision      : 25-bit
Duration       : 00:00:05.68 = 125184 samples ~ 425.796 CDDA sectors
File Size      : 501k
Bit Rate       : 706k
Sample Encoding: 32-bit Floating Point PCM

---

I have converted 1.wav to 1.mov and posted it here so you can listen to it.

1.mov

[csukuangfj]

TODOs

• Support writing wav to file
• Provide Python API
• Provide C API
• Provide Go API
• Provide C# API
• Provide Android demos
• Provide iOS demos
• Add documentation