- Home
- Text-to-speech tools
- F5-TTS
F5-TTS
What it is
F5-TTS is an open-source text-to-speech project that generates speech from text. It offers command-line inference and a Gradio web interface, with chunk inference, multiple styles and speakers, and voice chat powered by Qwen2.5-3B-Instruct. Its model lineup includes F5-TTS, described as using flow matching with a Diffusion Transformer and ConvNeXt V2, plus E2 TTS with a Flat-UNet Transformer. Custom inference guidance covers adding language support. You can install the project as a Python package, run an editable local clone, or use Docker. Training and fine-tuning work through Hugging Face Accelerate or a Gradio app. Installation guidance covers NVIDIA, AMD, Intel, and Apple Silicon hardware; AMD ROCm instructions specify Linux. Deployment documentation describes Triton and TensorRT-LLM. Python 3.10 or newer and FFmpeg are required. Code uses MIT, while pretrained models use CC-BY-NC. Commercial use is unavailable.
For and against
For
- Open-source MIT-licensed code
- Command-line and Gradio interfaces
- Multi-style and multi-speaker generation
- Training and fine-tuning support
- Docker installation option
Against
- Commercial use is unavailable
- Pretrained models use CC-BY-NC
- Requires Python 3.10 or newer and FFmpeg
- AMD ROCm guidance is Linux-specific
The verdict
F5-TTS provides several ways to run, customize, and deploy text-to-speech models. Licensing differs between code and pretrained models, and commercial use is not available.
Who it suits
F5-TTS suits developers and researchers building self-hosted speech systems, custom inference, or voice applications. It is relevant for users with supported GPU or Apple Silicon setups.
On the record
- API access
- Yesgithub.com · read 20 Sep 2026
- Commercial use
- Nogithub.com · read 20 Sep 2026
- Custom languages
- The README links to custom inference instructions for adding language support.github.com · read 8 Oct 2026
- Deployment
- The README describes a deployment solution using Triton and TensorRT-LLM.github.com · read 8 Oct 2026
- Hardware and operating systems
- The installation instructions cover NVIDIA, AMD, and Intel GPUs and Apple Silicon; the AMD ROCm instructions specify Linux.github.com · read 8 Oct 2026
- Hardware requirements
- The installation instructions cover NVIDIA, AMD, Intel, and Apple Silicon setups; the AMD ROCm instructions specify Linux.github.com · read 8 Oct 2026
- Inference features
- The Gradio app supports basic text-to-speech with chunk inference, multi-style and multi-speaker generation, and voice chat powered by Qwen2.5-3B-Instruct.github.com · read 8 Oct 2026
- Interfaces
- The project provides a Gradio web interface and command-line inference.github.com · read 8 Oct 2026
- License
- The code is released under the MIT License, while the pre-trained models are licensed under CC-BY-NC.github.com · read 8 Oct 2026
- Model availability
- The README links F5-TTS and E2 TTS base models on Hugging Face, Model Scope, and Wisemodel.github.com · read 8 Oct 2026
- Model design
- The project describes F5-TTS as a Diffusion Transformer with ConvNeXt V2 and includes E2 TTS, a Flat-UNet Transformer.github.com · read 8 Oct 2026
- Purpose
- F5-TTS is an open-source text-to-speech project that uses flow matching to generate fluent and faithful speech.github.com · read 8 Oct 2026
- Related tools
- The project acknowledges Vocos and BigVGAN as vocoders and FunASR and faster-whisper among its evaluation tools.github.com · read 8 Oct 2026
- Requirements
- The installation instructions call for Python 3.10 or newer and FFmpeg.github.com · read 8 Oct 2026
- Runtime benchmark
- The README reports a 253 ms average latency for F5-TTS Base with Vocos at concurrency 2 on a single L20 GPU in its stated benchmark setup.github.com · read 8 Oct 2026
- Support guidance
- The README recommends searching GitHub issues for help with problems encountered.github.com · read 8 Oct 2026
- Training
- Training and fine-tuning are supported through Hugging Face Accelerate or a Gradio app.github.com · read 8 Oct 2026
- Voice cloning
- Yesgithub.com · read 20 Sep 2026
