F5-TTS

What F5-TTS is

F5-TTS is an open-source text-to-speech project for generating speech from text, with command-line inference and a Gradio web interface. Its approach uses flow matching and a Diffusion Transformer with ConvNeXt V2; the project also includes E2 TTS, which uses a Flat-UNet Transformer. The Gradio app supports chunked inference, multiple styles and speakers, and voice chat powered by Qwen2.5-3B-Instruct. Custom inference instructions cover adding language support. Users can install it as a Python package, run it from an editable local clone, or build and run it with Docker. Training and fine-tuning are supported through Hugging Face Accelerate or a Gradio app. Installation guidance covers NVIDIA, AMD and Intel GPUs, as well as Apple Silicon; AMD ROCm instructions specify Linux. The project also describes deployment with Triton and TensorRT-LLM. Requirements include Python 3.10 or newer and FFmpeg. The code is under the MIT License, but pre-trained models use CC-BY-NC, and commercial use is listed as unavailable.