XTTS with Gradio for AI Training

This repository copies most of the files from Coqui-TTS, with some changes in order to create an image with correct dependencies for AI voice fine-tuning in Web UI.

Some of the changes include:

• The Dockerfile;
• Frozen pip dependencies: requirements.frozen.txt;
• ./TTS/demos/xtts_ft_demo/xtts_demo.py does not create a public link and disables analytics;
• ./TTS/demos/xtts_ft_demo/utils/formatter.py set to use Whisper's large-v3 model.

Build the Docker/Podman Image

Install Docker or Podman and adapt the commands accordingly:

podman build . -t xtts_gradio

Pulling the Image

Instead of building the image from source, you can pull it from docker.io or ghcr.io:

podman pull docker.io/lvxvvv/xtts_gradio:latest

podman pull ghcr.io/lvxvvv/xtts_gradio:latest

Running the Container

If you are using Podman, simply run:

./run.sh

The following command will be executed:

podman run -it --gpus all -p 5003:5003 --entrypoint python3 xtts_gradio /root/tts/TTS/demos/xtts_ft_demo/xtts_demo.py

If you are using docker, modify this file so it uses docker instead of podman.

This commands runs the container with an entrypoint that initializes the Web UI. It will be served on port 5003 on your localhost:

License

Since this repo's purpose is the creation of a working Web UI from the original repo, it will follow MPL-2.0, the same as in Coqui-TTS. For legal questions about the use of this tool, the user must refer to Coqui's page.