F5-TTS: Flow Matching Text-to-Speech

1.1.20 verified Fri May 01 auth: no python

F5-TTS is a text-to-speech library using flow matching, designed for faithful and fluent speech synthesis. Current version: 1.1.20. Active development with frequent releases.

pip install f5-tts

Common errors

error ModuleNotFoundError: No module named 'f5_tts' ↓

cause Library not installed or virtual environment not activated.

fix

pip install f5-tts

error ImportError: cannot import name 'F5TTS' from 'f5_tts' ↓

cause Outdated version (<1.1.0) where F5TTS was in a submodule.

fix

pip install --upgrade f5-tts, then use from f5_tts import F5TTS

error RuntimeError: MelSpectrogram cache issue ↓

cause Known bug in v1.1.18 fixed in v1.1.19.

fix

Upgrade to f5-tts>=1.1.19

Warnings

gotcha Gradio version conflict: Gradio >=6.11 causes UI freeze. Pin to <6.11 or use version from requirements. ↓

fix pip install 'gradio<6.11'

deprecated Direct import of model classes (e.g., `from f5_tts.model import F5TTS`) may break in future releases. Use top-level `from f5_tts import F5TTS`. ↓

fix from f5_tts import F5TTS

gotcha Inference requires both reference audio and reference text. Omitting ref_text may yield poor quality. ↓

fix Always provide matching reference text for best results.

Install

pip install f5-tts[infer]

Imports

F5TTS
wrong
```
from f5_tts.model import F5TTS
```
correct
```
from f5_tts import F5TTS
```
Old import path used in v1.0; main API moved to top-level f5_tts
infer_batch
wrong
```
from f5_tts import infer_batch
```
correct
```
from f5_tts.infer import infer_batch
```
infer_batch is not exposed at package level; must import from .infer module

Quickstart

Quick start for basic inference with a reference audio clip.

from f5_tts import F5TTS

tts = F5TTS()

# Generate speech from text
waveform, sample_rate = tts.infer(text="Hello, world!", ref_audio="path/to/ref.wav", ref_text="Reference text")

# Save to file
import soundfile as sf
sf.write("output.wav", waveform, sample_rate)