Repolode
Submit
SWivid

SWivid/F5-TTS

Open-source text-to-speech system using flow matching with a Diffusion Transformer, supporting zero-shot voice cloning from a short reference audio.

Open-source alternative to:ElevenLabs

View on GitHubWebsite
Stars
15,427
Forks
2,250
Open issues
65
Last commit
6 days ago

What is it?

F5-TTS is a text-to-speech system built on flow matching with a Diffusion Transformer (DiT) that uses ConvNeXt V2 blocks. It generates speech from text and a short reference audio clip, reproducing the reference speaker's voice without per-speaker training. The project also includes the E2 TTS variant, a flat-UNet Transformer that the authors describe as a close reproduction of the original paper, and a Sway Sampling inference strategy for flow steps.

The repository provides a pip package, a Gradio web interface, and a command-line tool driven by TOML configuration files for inference. It also includes training and finetuning workflows based on Hugging Face Accelerate, a Gradio finetuning app, Docker images, and a Triton and TensorRT-LLM runtime for deployment.

Pre-trained models are released under CC-BY-NC because the training data, Emilia, is in-the-wild data, while the code itself is MIT licensed.

Key features

  • 01Zero-shot voice cloning from a short reference audio and its transcript.
  • 02Gradio web interface for inference and a finetuning app.
  • 03Command-line inference driven by TOML config files, including multi-voice stories.
  • 04Training and finetuning with Hugging Face Accelerate.
  • 05Docker images and a Triton plus TensorRT-LLM runtime for deployment.

Use cases

  • Generating narration or voiceovers in a cloned voice from a short sample.
  • Researching flow-matching text-to-speech models and their training recipes.
  • Deploying a self-hosted TTS service with GPU acceleration via Triton and TensorRT-LLM.
  • Finetuning a base model for a new language or speaker.

Quick start

  1. 1

    Create and activate a Python 3.10+ environment

    conda create -n f5-tts python=3.11
  2. 2

    Install PyTorch matching your device (e.g. NVIDIA CUDA build)

    pip install torch==2.8.0+cu128 torchaudio==2.8.0+cu128 --extra-index-url https://download.pytorch.org/whl/cu128
  3. 3

    Install the F5-TTS pip package for inference

    pip install f5-tts
  4. 4

    Launch the Gradio web interface

    f5-tts_infer-gradio

Things to know

  • Pre-trained models are CC-BY-NC licensed, so they cannot be used commercially.
  • Requires a GPU for practical inference speed.
  • Requires FFmpeg and a matching PyTorch install for your hardware.

Topics

#text-to-speech#speech-synthesis#flow-matching#diffusion-transformer#voice-cloning#pytorch#deep-learning

FAQ

What is F5-TTS?
Open-source text-to-speech system using flow matching with a Diffusion Transformer, supporting zero-shot voice cloning from a short reference audio. F5-TTS is a text-to-speech system built on flow matching with a Diffusion Transformer (DiT) that uses ConvNeXt V2 blocks. It generates speech from text and a short reference audio clip, reproducing the reference speaker's voice without per-speaker training. The project also includes the E2 TTS variant, a flat-UNet Transformer that the authors describe as a close reproduction of the original paper, and a Sway Sampling inference strategy for flow steps.
Is F5-TTS free to use? What license does it have?
F5-TTS is open source under the MIT license. Check the license terms before using it in commercial products.
Is F5-TTS still maintained?
Actively maintained. The last commit to F5-TTS was on Oct 14, 2026.
What can I use F5-TTS for?
Generating narration or voiceovers in a cloned voice from a short sample. Researching flow-matching text-to-speech models and their training recipes. Deploying a self-hosted TTS service with GPU acceleration via Triton and TensorRT-LLM. Finetuning a base model for a new language or speaker.
How do I install F5-TTS?
Create and activate a Python 3.10+ environment: conda create -n f5-tts python=3.11 Install PyTorch matching your device (e.g. NVIDIA CUDA build): pip install torch==2.8.0+cu128 torchaudio==2.8.0+cu128 --extra-index-url https://download.pytorch.org/whl/cu128 Install the F5-TTS pip package for inference: pip install f5-tts Launch the Gradio web interface: f5-tts_infer-gradio
Where can I get F5-TTS?
The source code and installation instructions are on GitHub: https://github.com/SWivid/F5-TTS

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects