This repo is a SG2300x-adapted demo of EmotiVoice(@NetEase Youdao) and OpenVoice(@MyShell.ai).

• Download the bmodels for EmotiVoice.

python3 -m pip install dfn
python3 -m dfn --url https://disk.sophgo.vip/sharing/KymDuWLGw
unzip EmotiVoice.zip
mv EmotiVoice model_file/tts
rm EmotiVoice.zip

if you have no access to sophgo disk, you can use baidu disk instead.

• Download bmodels of OpenVoice tone color converter here and run tar zxfv checkpoints.tar.gz, mv checkpoints/converter model_file/converter, rm -rf checkpoints.tar.gz checkpoints.

• The model_file directory should be organized as:

model_file
├── converter
│   ├── checkpoint.pth
│   ├── config.json
│   ├── decoder_1024_f16.bmodel
│   └── decoder_2048_f16.bmodel
├── simbert-base-chinese
│   ├── config.json
│   └── vocab.txt
└── tts
    ├── am_durationpred-1_512_384-1_512_1.bmodel
    ├── am_energypred-1_512_384-1_512_1.bmodel
    ├── am_pitchpred-1_512_384-1_512_1.bmodel
    ├── am_rest_weight.pth
    ├── bert_poolout1-768_1-512_1-512_1-512.bmodel
    ├── hifigan_1-80-1024_F16.bmodel
    ├── jit_am_encoder_1-512-384_1-1-512.bmodel
    └── onnx_am_decoder_1-2048-384.bmodel

sudo apt-get install libsndfile1 -y
pip3 install torch torchaudio
pip3 install numpy numba scipy transformers==4.26.1 librosa soundfile yacs g2p_en jieba pypinyin whisper_timestamped onnxruntime gradio==4.19.2
mv assets/master.zip ~/.cache/torch/hub/
cd ~/.cache/torch/hub/
unzip master.zip
rm master.zip
mv snakers4-silero-vad-6c8d844 snakers4_silero-vad_master
echo "export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libgomp.so.1" >> ~/.bashrc
source ~/.bashrc
cd -
python3 demo_page.py

• Install the dependencies: pip3 install fastapi pydub uvicorn[standard] pyrubberband.
• Then, run uvicorn openai_api:app --reload --host 0.0.0.0 --port [port_number], the service will be available at hostip:port_number in few seconds. The way to call the service is exactly the same as using OpenAI's TTS service.