VoiceStudio Clones, Designs and Dubs Voices in 646 Languages on Your Own Computer as an Open-Source ElevenLabs Alternative

The feature list reads like a paid voice platform’s pricing page. Cloning makes a new voice from a sample. Voice design skips the sample entirely: “Describe a voice in words and generate it.” Dubbing goes further: “Translate and dub videos with speech timed to the original.” Dictation adds “a floating dictation widget and local transcription,” and the audiobook tool narrates long text.

Under the hood, developer Palash Debnath leans on other open projects. Speech generation is “powered by k2-fsa/OmniVoice,” speech recognition comes from whisper.cpp, and CosyVoice models run in the GGUF format that local-AI tools use to fit models on ordinary hardware.

It runs on macOS (Apple Silicon), Windows 10 and 11, and Linux, with Android 10+ and iOS 17+ listed as previews. The README’s answer to the first question anyone asks about local AI: “No GPU? The app is fully usable on the CPU, just slower.” Models are big, so “the app asks before it downloads any model.” For developers there’s a local API and an MCP server, so an AI agent on the same machine can call the voices too.

The license is AGPL-3.0, which allows anyone to use and modify it but requires sharing the source of a modified version run as a network service. The README’s ethics rule fits in one line: “Clone a voice only with its owner’s permission.” The app can’t check that for you, and the README doesn’t say how, or whether, it marks generated audio as synthetic.

The 94,000 stars in about six months show how much demand there was for this to run locally. The same thing that makes it appealing, no server in the loop, makes it a harder problem for anyone worried about voice fraud: the safeguards a hosted service can enforce become a sentence in a README. Watch for whether the project adds watermarking or a consent step, and how the phone previews hold up on the CPU.