F5-TTS
SWivid/F5-TTS · Python · MIT
Fast zero-shot TTS with natural prosody
F5-TTS became a favorite for local voice cloning demos: strong quality, reasonable speed, and an MIT license that is friendly for experiments.
Score
Setup at a glance
pip install f5-tts Deep dive
F5-TTS is a zero-shot text-to-speech system that clones a voice from a short reference clip — no fine-tuning per speaker — using a flow-matching model trained mostly on the Emilia dataset.
The v1 base model (March 2025) brought better training and inference code, and the same family includes the E2 TTS base models; weights are published on Hugging Face, ModelScope and WiseModel so you can pick a mirror.
It runs as a Gradio WebUI, a CLI and a Docker image, with an updates blog tracking releases and a paper explaining the flow-matching approach.