English · Русский
VoiceBridge turns speech into text on your own computer and inserts the result into the active application. Hold a configured mouse button or keyboard key, speak, release and get text without using a hosted speech-recognition API.
It is a free, open-source desktop app built around locally executed Qwen3-ASR inference. The initial installation and model download may require internet access; recognition using a downloaded model is local.
- Push-to-talk from a mouse button or keyboard shortcut (for example,
F8as configured). - Local transcription with Qwen3-ASR; select an existing model or download one separately.
- Inserts recognized text via clipboard/keyboard integration, with platform-specific controls.
- Desktop Qt interface, personal vocabulary, text settings, model management and diagnostics.
- Supports both CPU and compatible GPU environments, depending on installed inference dependencies and hardware.
| Operating system | Version | Package | Status |
|---|---|---|---|
| Ubuntu/Debian amd64 | 0.5.6 | VoiceBridge-Ubuntu-v0.5.6-Apache-2.0.zip |
Desktop + DEB + Python wheel |
| Windows 10/11 x64 | 0.5.4 | VoiceBridge-Windows-v0.5.4-Apache-2.0.zip |
Online EXE installer; unsigned |
These are two separately maintained release snapshots, not a single unified cross-platform 0.5.6 release. See the complete, version-specific sources under platforms/ubuntu-v0.5.6 and platforms/windows-v0.5.4.
-
Download the Ubuntu 0.5.6 release ZIP from GitHub Releases and fully extract it.
-
In the extracted directory, as your regular desktop user, run:
bash install.sh voice-bridge-desktop
-
Choose an existing Qwen3-ASR model or download one from the application.
Do not run the install script as sudo bash install.sh; it requests elevated privileges only for required steps. Optional diagnostics: bash check.sh and voice-bridge-desktop --diagnostics. The provided public bundle omits historical rollback packages; upgrading an old installation may require a separate migration procedure. Do not remove a working older installation solely to try this bundle.
- Download the Windows 0.5.4 release ZIP and fully extract it.
- Run
VoiceBridge-Setup-0.5.4-x64.exeas a normal user. - Allow the installer to obtain missing Python and AI packages, then choose or download an ASR model.
The installer is not Authenticode-signed. It is an online installer, not a fully offline package. The public package does not include private migration/rollback files for older installations.
Python 3.11+ is required. Each version has its own Python package, pyproject.toml, tests and platform packaging scripts:
# Ubuntu source snapshot
cd platforms/ubuntu-v0.5.6
python packaging/build.py
# Windows source snapshot (requires LLVM clang/lld-link)
cd platforms/windows-v0.5.4
python packaging/windows/build_release.py --out /path/outside/sourceThe two commands are separate; run each from its own source directory. The Windows command builds an installer but does not itself validate behavior on a Windows machine.
- Transcription runs locally; the source does not require a hosted ASR service.
- Internet access is used when installing dependencies or downloading a model, including packages from third-party hosts.
- No model weights, microphone recordings, machine-specific configurations, private keys or internal infrastructure are included in the published source.
- VoiceBridge interacts with the clipboard and desktop input APIs. Review permissions, downloaded software and target application privacy policies before use.
VoiceBridge's original source code: Apache License 2.0 · Copyright 2025–2026 Andrey154154.
Third-party packages and models retain their own licenses. The Qwen3-ASR model is not redistributed by this project. See third-party notices and NOTICE. The Apache license does not grant trademark rights.
For installation specifics and platform caveats, see Ubuntu release notes and Windows release notes.