English | 简体中文
An offline voice-command tool for Windows. Say a short word and it sends a keystroke, a key combination, or a macro. Built for users who need an alternative input method.
Recognition runs entirely on-device with no network requests; commands, recordings and templates stay local and are stored encrypted.
- Custom recordings — record a few takes per command and it starts triggering
- Single-word triggering — character-level discrimination, with multi-character commands and multiple pronunciations per command
- Continuous triggering — keep speaking commands back-to-back without restarting recognition
- Flexible key behaviour — tap, hold-to-toggle, macro sequences, mutually exclusive groups, command groups, global hotkey or voice toggle
Building requires Rust stable (MSVC). Running requires Windows 10/11 and administrator privileges.
cargo build --release
powershell -ExecutionPolicy Bypass -File packaging/package_dist.ps1
dist\Yuke-<version>\yuke.exe- The packaging script builds
yuke.pak(standard and noise-robust encoders; Silero VAD is bundled) and assembles the distribution directory. Runtime dependencies such asonnxruntime.dllare taken frompackaging/redist/. - Tests must run serially, as they use the audio device and temporary directories:
cargo test --all-targets -- --test-threads=1. - For development the app can also run unpackaged:
xcopy /E /I assets\model target\release\modelandcopy packaging\redist\*.dll target\release\(the pak takes precedence over the loose model files). - Pushing a tag such as
1.2.6triggers CI to build and publish a release.
- Settings → Audio: select a microphone.
- Commands → "+ New command": give it a name, add three takes, regenerate the pronunciation template, then capture the key.
- Back on the home page, press "Start recognition" and say the word.
- Architecture, data flow and CLI diagnostics: docs/ARCHITECTURE.md
- Data handling: PRIVACY.md
MIT — see LICENSE. Sources and licences of models and assets: ATTRIBUTIONS.md.