Support vits models from piper

[csukuangfj]

Usage

I have converted the model en_US-lessac-medium from piper to sherpa-onnx
and you can find the converted model in the following repo:
https://huggingface.co/csukuangfj/vits-piper-en_US-lessac-medium

If you are interested in how the model is converted, please have a look at the following colab notebook:
https://colab.research.google.com/drive/1PScLJV3sbUUAOiptLO7Ixlzh9XnWWoYZ?usp=sharing

In the following, we describe how to use the converted model with sherpa-onnx

Download models

wget https://huggingface.co/csukuangfj/vits-piper-en_US-lessac-medium/resolve/main/en_US-lessac-medium.onnx

wget https://huggingface.co/csukuangfj/vits-piper-en_US-lessac-medium/resolve/main/lexicon.txt

wget https://huggingface.co/csukuangfj/vits-piper-en_US-lessac-medium/resolve/main/tokens.txt

Build sherpa-onnx

Please use the code from this PR and build sherpa-onnx from source by following
https://k2-fsa.github.io/sherpa/onnx/install/index.html

After building sherpa-onnx, you can use

./build/bin/sherpa-onnx-offline-tts \
  --output-filename=./piper.wav \
  --vits-length-scale=0.8 \
  --vits-model=./vits-piper-en_US-lessac-medium/en_US-lessac-medium.onnx \
  --debug=1 \
  --vits-lexicon=./vits-piper-en_US-lessac-medium/lexicon.txt \
  --vits-tokens=./vits-piper-en_US-lessac-medium/tokens.txt \
  "Liliana, the most beautiful and lovely assistant of our team!"

It will generate a wave file ./piper.wav. I have converted it to mov and attached it below so that you can listen to it.

piper.mov

Your help is needed

Since piper provides a lot of pre-trained onnx models, it would be great if you can help convert them to sherpa-onnx.

Unlike piper, we don't need to install piper_phonemize. We have converted all the words into lexicon.txt.

[csukuangfj]

Note that you can find all the pre-trained ONNX models from piper at the following address
https://huggingface.co/rhasspy/piper-voices/tree/main