OpenVoice is an open-source instant voice cloning model with 38,000+ stars on GitHub, created by researchers from MIT and MyShell, that copies a voice from a short reference clip.
OpenVoice separates two things most cloning systems tangle together: the tone color of a voice — its recognizable timbre — and the style of a delivery. Clone the tone color from a brief reference recording, then independently control emotion, accent, rhythm, pauses and intonation in the generated speech. Because the two are decoupled, cloning works zero-shot across languages: neither the language of the reference nor of the generated speech needs to appear in the training data. Version 2, released in April 2024, improved audio quality, added native support for English, Spanish, French, Chinese, Japanese and Korean, and moved both versions to the MIT license for free commercial use. OpenVoice has powered MyShell’s instant voice cloning in production since 2023, where it was used tens of millions of times. The repository was most recently updated in 2025 and remains a go-to open reference for voice cloning.
OpenVoice is free and open source under the MIT license, explicitly including commercial use. You self-host the model, so the only cost is your own GPU compute.
Developers adding voice cloning to apps, games and content tools, and researchers who need a strong, permissively licensed baseline for speech generation.