Text-to-speech and tone color conversion on your Airbox

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

0. Clone the repository.

git clone https://github.com/ZillaRU/EmotiVoice-TPU.git

1. Download bmodels

bash download.sh

2. Setup environment

bash prepare.sh

2. Run web demo

Run the following script.

python3 demo_page.py

Three tabs for TTS, TTS+tone color cloning and tone color cloning are shown.

3. Using OpenAi-like api

• 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.

For more detailed usage, please refer to the README from the original repo of EmotiVoice and OpenVoice.

Issues

The torch version higher than 2.3 might lead to an ERROR due to poor compatibility for ARM platform.