source: kdnuggets: getting started with omnivoice-studio
level: technical
omnivoice studio is a desktop application that performs voice cloning, video dubbing, real-time dictation, and voice design entirely on your local machine. it is free for personal use and requires no api keys, accounts, or subscriptions. the app supports 646 languages and uses a zero-shot diffusion text-to-speech model that clones a voice from a three-second audio clip. it is built with tauri, react, and fastapi, and relies on open-source components like whisperx for transcription, demucs for vocal isolation, omnivoice for synthesis, and pyannote for speaker diarization.
the app auto-detects gpu hardware and can run on cpu if needed, though synthesis is slower. system requirements include 8 gb ram, 4 gb vram, and 10 gb disk space. installation involves cloning the github repository, installing dependencies with bun, and launching via bun run desktop-prod. pre-built installers are available for macos, windows, and linux. a hugging face token is optional but needed for speaker diarization and some voice-design features. the first launch downloads about 2.4 gb of model weights.
voice cloning works by providing a three-to-ten-second reference clip and selecting a target language. the dubbing pipeline transcribes a video, translates the speech, clones the original voices, synthesizes dubbed audio, and mixes it back into the video—all locally. the app also includes vocal isolation to clean reference clips. output can be exported as mp3, wav, or flac. the project is in active beta, with the latest release being v0.2.7.
why it matters: it provides a privacy-preserving, cost-free alternative to cloud-based voice ai services, enabling sensitive or proprietary audio work without data leaving the user's machine.