Clone any voice with just one minute of audio
Free alternative to ElevenLabs
GPT-SoVITS lets you clone voices and generate natural speech from text using only short audio samples. A 5-second clip enables instant voice synthesis, while 1 minute of audio allows fine-tuning for much higher realism. It features a browser-based interface and supports cross-lingual generation in English, Chinese, Japanese, Korean, and Cantonese.
Local training requires a capable GPU; Mac users are limited to CPU training.
Not sure how to install, or if it’s safe? Read the beginner guide
Yes, GPT-SoVITS is completely free and open source under the MIT license, with no API keys or subscriptions required to run it.
It is open source (MIT license) with 64,987 GitHub stars, last updated 2026-11-11. Download only from the official link on this page, and scan the file before installing: see the safety checklist in our beginner guide.
Yes, it supports cross-lingual voice cloning and text-to-speech across English, Chinese, Japanese, Korean, and Cantonese from short audio samples.
Compiled from the project README. 64,987 GitHub stars (2026-11-13) · License MIT · Last update 2026-11-11