# F5-TTS review

> F5-TTS is an open-source Python text-to-speech system using flow matching with a Diffusion Transformer, supporting zero-shot voice cloning from reference audio.

Devtoolwatch Board, the week of Oct 25–31, 2026. Canonical page: https://index.devtoolwatch.com/tools/f5-tts/

## TL;DR

F5-TTS is the official code for a non-autoregressive text-to-speech model based on flow matching and a Diffusion Transformer. F5-TTS is for developers who want to generate speech from a reference audio clip, locally or with Docker. F5-TTS adds Sway Sampling at inference time and includes a Gradio app, a CLI, and finetuning tools.

## F5-TTS at a glance

- **Category:** [Voice and realtime](https://index.devtoolwatch.com/categories/voice-and-realtime/)
- **Maintained by:** Yushen CHEN (individual, [@SWivid](https://github.com/SWivid) on GitHub)
- **License:** MIT
- **Language:** Python, Shell, Dockerfile
- **Repository:** [github.com/SWivid/F5-TTS](https://github.com/SWivid/F5-TTS)
- **Website:** [arxiv.org/abs/2410.06885](https://arxiv.org/abs/2410.06885)
- **PyPI package:** [f5-tts](https://pypi.org/project/f5-tts/) 1.1.22
- **Latest release:** [1.1.22](https://github.com/SWivid/F5-TTS/releases/tag/1.1.22) on GitHub, Jul 23, 2026
- **First published:** Oct 8, 2024

## What F5-TTS does

- F5-TTS generates speech from text, using a reference audio clip and its transcript to set the voice.
- F5-TTS offers a Gradio web app with chunk inference, multi-style and multi-speaker generation, and voice chat powered by Qwen2.5-3B-Instruct.
- F5-TTS provides a CLI that accepts flags or a .toml config file, including multi-voice story configs.
- F5-TTS can transcribe the reference audio with an ASR model when ref_text is left empty, at the cost of extra GPU memory.
- F5-TTS supports training and finetuning through Hugging Face Accelerate or a finetuning Gradio app.
- F5-TTS includes a deployment option using Triton and TensorRT-LLM.
- F5-TTS also ships an E2 TTS implementation, described as the closest reproduction of the E2 TTS paper.

### How it works

F5-TTS runs as a Python package on top of PyTorch, installed for your GPU type: NVIDIA CUDA, AMD ROCm, Intel XPU, or Apple Silicon. You run inference through the f5-tts_infer-gradio web app or the f5-tts_infer-cli command. Pre-trained base models are published on Hugging Face, Model Scope, and Wisemodel. Docker images on GitHub Container Registry serve the web interface on port 7860.

## How to install F5-TTS

**With pip (inference only)**

```
pip install f5-tts
```

**With Docker (web interface)**

```
docker container run --rm -it --gpus=all --mount 'type=volume,source=f5-tts,target=/root/.cache/huggingface/hub/' -p 7860:7860 ghcr.io/swivid/f5-tts:main f5-tts_infer-gradio --host 0.0.0.0
```

### Requirements

- Python 3.10 or later
- FFmpeg
- PyTorch and torchaudio built for your device (CUDA, ROCm on Linux, XPU, or Apple Silicon)
- RDNA 3.5 and RDNA 4 AMD GPUs require ROCm 7.x

### First steps

1. Create a Python 3.10 or later environment, install FFmpeg, and install PyTorch matched to your GPU.
2. Install the package with pip install f5-tts if you only need inference.
3. Launch the web interface with f5-tts_infer-gradio, or run f5-tts_infer-cli with the default settings.
4. Pass --ref_audio, --ref_text, and --gen_text to the CLI with --model F5TTS_v1_Base to generate speech in a chosen voice.

## Who F5-TTS is for

- Developers running zero-shot voice cloning text-to-speech on their own GPU
- Researchers training or finetuning flow-matching TTS models
- Teams serving TTS with Triton and TensorRT-LLM
- Non-commercial projects that can accept the CC-BY-NC model license

### Limitations

- F5-TTS pre-trained models use the CC-BY-NC license because of the Emilia training data, though the code is MIT.
- F5-TTS AMD GPU support through ROCm is Linux only.
- F5-TTS uses extra GPU memory when ref_text is empty, because an ASR model transcribes the reference audio.
- F5-TTS Intel XPU install requires Intel Deep Learning Essentials or oneAPI Base Toolkit, unless you use IPEX.

## Usage data (the week of Oct 25–31, 2026)

- Board rank: #188 overall, #11 in Voice and realtime, momentum score 46
- Downloads that week: 25.1k (+56% vs 4-week average)
- New GitHub stars per week (last 12 weeks): Aug 9: 44; Aug 16: 33; Aug 23: 25; Aug 30: 33; Sep 6: 24; Sep 13: 42; Sep 20: 37; Sep 27: 51; Oct 4: 40; Oct 11: 55; Oct 18: 41; Oct 25: 53

## F5-TTS alternatives

| Rank | Tool | Stars | Downloads/wk |
|---|---|---|---|
| #1 | [faster-whisper](https://index.devtoolwatch.com/tools/faster-whisper/) | 26.4k | 4.4M |
| #2 | [Kokoro](https://index.devtoolwatch.com/tools/kokoro/) | 9.8k | 150.7k |
| #3 | [Chatterbox](https://index.devtoolwatch.com/tools/chatterbox/) | 27.5k | 73.2k |
| #4 | [VoiceMem](https://index.devtoolwatch.com/tools/xzf-thu-voicemem/) | 2.8k | – |
| #5 | [Pipecat](https://index.devtoolwatch.com/tools/pipecat/) | 17k | 348.5k |
| #6 | [WhisperX](https://index.devtoolwatch.com/tools/whisperx/) | 25k | 158.6k |

## F5-TTS FAQ

### Is F5-TTS free and open source?

Yes. F5-TTS's source code is on GitHub (SWivid/F5-TTS) under the MIT license, so you can use it for free.

### How do I install F5-TTS?

Run `pip install f5-tts`. The install section above has the other options from the project's README.

### Is F5-TTS still maintained?

Yes. The SWivid/F5-TTS repository was last updated on Oct 14, 2026, and the latest release, 1.1.22 on GitHub, came out on Jul 23, 2026.

### How popular is F5-TTS?

F5-TTS has 15.5k GitHub stars as of Nov 1, 2026. It gained 53 new stars in the week of Oct 25–31, 2026, against a 12-week average of 40 a week. Its main package had 25.1k downloads that week, +56% against its own 4-week average. That made it #11 among voice and realtime AI tools on the Devtoolwatch Board.

### F5-TTS vs Chatterbox: how do they compare?

F5-TTS and Chatterbox are both tracked by Devtoolwatch as voice and realtime AI tools. F5-TTS has 15.5k GitHub stars and Chatterbox has 27.5k. In the week of Oct 25–31, 2026, F5-TTS's main package had 25.1k downloads and Chatterbox's had 73.2k. Chatterbox is a family of open-source text-to-speech and voice cloning models by Resemble AI, with English and multilingual variants. [Chatterbox review](https://index.devtoolwatch.com/tools/chatterbox/)

### What are the best alternatives to F5-TTS?

Other voice and realtime AI tools on the Devtoolwatch Board in the week of Oct 25–31, 2026, by momentum: faster-whisper (#1), Kokoro (#2), Chatterbox (#3), VoiceMem (#4), Pipecat (#5). The alternatives table above has their numbers.

### How do I use F5-TTS locally?

F5-TTS runs locally after you install PyTorch for your GPU and run pip install f5-tts. Then start the web interface with f5-tts_infer-gradio, or use f5-tts_infer-cli with a reference audio file, its transcript, and the text to generate.

### Can I use F5-TTS models commercially?

F5-TTS code is MIT licensed. The pre-trained models are licensed CC-BY-NC because they were trained on the Emilia in-the-wild dataset, which restricts commercial use of those checkpoints.

### Which GPUs does F5-TTS support?

F5-TTS supports NVIDIA GPUs with CUDA, AMD GPUs with ROCm on Linux, Intel GPUs through XPU or IPEX, and Apple Silicon with the stable PyTorch build.

### Do I need to transcribe the reference audio for F5-TTS?

F5-TTS does not require a transcript. If you leave --ref_text empty, an ASR model transcribes the reference audio, which uses extra GPU memory.

### Can I finetune F5-TTS on my own data?

F5-TTS supports training and finetuning with Hugging Face Accelerate or through a Gradio app launched with f5-tts_finetune-gradio. Use the local editable install for training.

## Sources

1. [SWivid/F5-TTS on GitHub](https://github.com/SWivid/F5-TTS)
2. [F5-TTS README](https://github.com/SWivid/F5-TTS/blob/main/README.md)
3. [f5-tts on PyPI](https://pypi.org/project/f5-tts/)
4. [F5-TTS website](https://arxiv.org/abs/2410.06885)
5. [Release 1.1.22 (Jul 23, 2026)](https://github.com/SWivid/F5-TTS/releases/tag/1.1.22)
6. [Release 1.1.21 (Jul 5, 2026)](https://github.com/SWivid/F5-TTS/releases/tag/1.1.21)
7. [Release 1.1.20 (Apr 20, 2026)](https://github.com/SWivid/F5-TTS/releases/tag/1.1.20)
8. [Hacker News: Successful room temperature ambient-pressure magnetic levitation of LK-99](https://news.ycombinator.com/item?id=36994214)
9. [Hacker News: The first room-temperature ambient-pressure superconductor?](https://news.ycombinator.com/item?id=36864624)
10. [Hacker News: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL](https://news.ycombinator.com/item?id=42823568)
11. [Devtoolwatch Board, the week of Oct 25–31, 2026](https://index.devtoolwatch.com/)