Custom Gradio F5-TTS

This project provides a Dockerised Gradio application for the F5-TTS voice cloning model. It extends the standard Gradio server with additional API endpoints to control model loading and unloading and to generate multiple clips from a single reference voice. The web UI exposes buttons for loading and unloading the model as well as a tab for generating several clips in one go.

Features

• /load – load the F5-TTS model into VRAM.
• /unload – remove the model from VRAM while keeping cache files on disk.
• /several – supply a reference voice and newline separated phrases to generate several WAV files in one request.

Running locally

python __main__.py

Docker

docker build -t custom-f5-tts .
docker run -p --gpus all 7860:7860 custom-f5-tts

The Gradio UI will be available at http://localhost:7860/. The additional API routes can be accessed on the same host.