OmniVoice-Studio

OmniVoice Studio is a local, open source answer to ElevenLabs. It clones a voice from a three-second clip, does text-to-speech across 14 engines, dubs video (transcribe, translate, re-voice, export MP4), builds audiobooks from EPUB or PDF, and runs a system-wide dictation widget. All of it runs on your own machine, AGPL-3.0 licensed, no subscription.

Running it well needs a real GPU. The minimum is 8GB RAM and CPU-only mode, but the recommended setup is 16GB-plus RAM and 8GB-plus of VRAM, with support for NVIDIA CUDA, Apple Silicon, and AMD ROCm. Prebuilt installers exist for macOS, Windows, and Linux, plus Docker, so getting it running is not the hard part; having hardware that makes it fast is.

For a solo creator or a small team doing voiceover, dubbing, or audiobook work, the fact that this is free and local is the whole pitch. No per-character billing, no clips leaving your machine, which matters when you are cloning voices. Larger teams can run it too, but the AGPL license means if you build it into a hosted product, you owe source; a commercial license is coming for that case.

The catch is voice cloning ethics and quality. Cloning a voice from three seconds is powerful and easy to misuse, and that is on you to handle responsibly. On quality, local open models are close to the commercial services but not always ahead of them, so test against your actual use case before you cancel the paid tool.