Install
OpenVoice runs on Windows 10 and 11. One .exe, with the speech engine inside it, so there is no Python to install and nothing to configure.
Get the installer
Download the installer — OpenVoice_<version>_x64-setup.exe — from the download page, which always offers the current release, or straight from the releases page. Then run it.
You will need roughly 700 MB of disk. That is the app plus the speech model it includes, so there is nothing further to fetch; the two optional models are extra only if you choose them.
Windows will warn you
SmartScreen shows a blue dialog about an unknown publisher, and the button you want is hidden behind More info, then Run anyway.
The installer is not code-signed. A certificate is wanted but unscheduled. Because you are being asked to click past a security warning, every release ships a SHA-256 beside the installer and is built by a public GitHub Actions run from the tagged commit, so you can check that what you downloaded is what CI built:
Get-FileHash .\OpenVoice_<version>_x64-setup.exe -Algorithm SHA256
Compare the result with the hash in the release notes and in the .sha256 file published next to the installer. If they differ, do not run it.
The download page carries the same command with the current filename already filled in, and the published hash beside it, so you can copy both from there rather than typing a version by hand.
Two signatures are involved and they are worth keeping apart. The installer is unsigned, which is what SmartScreen is objecting to. Separately, each release is signed with a minisign key whose public half is compiled into the app, so an update that fails verification is discarded without running. That protects upgrades; it does not stop the first install being flagged.
The first launch
Nothing is downloaded. The speech model is inside the installer, so the first sentence works with the network off, and so does every one after it.
Two things can reach the network, and you start both. One is an optional extra model — a multilingual one, or a smaller one for a machine short on memory — fetched only when you press Download on the Speech model screen. The other is the update check, which you can switch off in Settings → Updates; switching it off means no request is made at all.
Then
Hold Right Ctrl, speak, and let go. See first dictation.
Uninstalling
Use Windows Add or remove programs. Your history at %APPDATA%\OpenVoice\history.db and your downloaded models under %APPDATA%\OpenVoice\models are yours and are left alone. Delete that folder if you want them gone too.
Building from source
You do not need to, but you can, and it is the only way to use a GPU today. The sequence is in CONTRIBUTING.md. Expect roughly 16 GB of toolchain between Rust, the MSVC build tools and a Python environment.
The parts most worth hacking on need none of that. ov-core and ov-format are pure Rust that build and test on any platform:
cargo test -p ov-core -p ov-format