myshell-ai/openvoice is a voice generation tool that clones the timbre of a reference voice and produces synthesized speech in that voice. It is written in Python and released under the MIT license. The repository was created on November 29, 2023 and has gathered 38,019 stars. There was no release tag to confirm this time, so this is closer to a finding that points at the project itself rather than a new version.
It is a project made jointly by MIT and MyShell, and the repository description presents the tool as an audio foundation model, that is, a base model that creates new sounds out of the sounds it has learned. The README explains that OpenVoice clones the timbre of a reference voice to generate speech in multiple languages and accents. It allows fine-grained control of voice styles such as emotion, accent, rhythm, pauses and pitch, and it supports zero-shot voice cloning — cloning a voice heard for the first time without separate training — even when the languages of the reference and generated speech are not included in the multilingual training data. It further states that OpenVoice V2 offers better audio quality and native support for English, Spanish, French, Chinese, Japanese and Korean, and that both V1 and V2 can be used free of charge for commercial and research purposes under the MIT license.
However, the information confirmed this time contains no figures or benchmarks showing how much better V2 is than the previous version. No release history remains either, so the extent of the improvement from V1 to V2 cannot be compared directly, and audio quality and native language support can be known only from the README description. These are not independently measured results, and the only remaining basis for judgement is the momentum score of 5.4. It makes sense to start with V2, where Korean is explicitly listed as native.