OpenSpeak

OpenSpeak is a local-first desktop dictation app built with Tauri, Rust, and whisper.cpp.

It is designed for fast global dictation workflows:

• Trigger recording from a global hotkey
• Speak naturally
• Stop and insert text where you are working

Features

• Local transcription via whisper.cpp (whisper-rs)
• Global hotkey toggle for start/stop dictation
• Menu bar (tray-first) app flow on macOS
• Recording overlay HUD for background visual feedback
• Two output modes:
  • clipboard: copy text for manual paste
  • auto-paste: copy then trigger paste automatically
• Basic spoken formatting commands:
  • comma, period, question mark
  • new line, new paragraph
• Model download and local model management:
  • Tiny (75 MB, fastest)
  • Base (142 MB)
  • Small (466 MB, recommended default)
  • Medium (1.5 GB)
  • Large-v3 (3.1 GB, highest accuracy)
  • Turbo (fast large model)

Architecture

• Frontend: React + Vite
• Desktop shell: Tauri v2
• Core runtime: Rust
• Audio capture: cpal
• Inference: whisper-rs / whisper.cpp

Requirements

• macOS (current focus)
• Node.js 20+
• Rust toolchain (rustup, cargo)
• Xcode Command Line Tools
• cmake (required by whisper build)

Install cmake:

brew install cmake

Getting Started

Install dependencies:

npm install

Run in development:

cargo tauri dev

Production build:

cargo tauri build

Build outputs:

• .app: src-tauri/target/release/bundle/macos/

Install (GitHub Releases, macOS)

Current release artifacts are unsigned macOS app bundles (.app.tar.gz).

1. Download the latest OpenSpeak_*.app.tar.gz from GitHub Releases.
2. Extract it and move OpenSpeak.app to /Applications.
3. Remove the quarantine flag once:

xattr -rd com.apple.quarantine /Applications/OpenSpeak.app

4. Launch OpenSpeak from Applications.

If macOS still blocks launch, use Finder Right-click -> Open once and confirm.

Permissions (macOS)

For full functionality, OpenSpeak needs:

• Microphone access (record speech)
• Accessibility access (auto-paste automation)

If auto-paste is disabled, Accessibility permission is optional.

Model Storage

By default, model files are stored under:

~/Library/Application Support/openspeak/models/

OpenSpeak also supports legacy paths from earlier project naming and will continue to read existing local data if present.

Configuration

OpenSpeak persists settings in local app data, including:

• Global hotkey
• Default model
• Paste mode
• Privacy flags

Default hotkey:

CommandOrControl+Shift+Space

Development Notes

• The app runs tray-first by default; open settings from the tray menu.
• Closing the settings window hides it instead of quitting.
• The overlay is a separate transparent always-on-top window.

GitHub Releases

This repository includes a GitHub Actions release workflow:

• File: .github/workflows/release.yml
• Trigger: push a semantic version tag (v*.*.*)
• Runner: macos-latest
• Artifacts uploaded to GitHub Releases via tauri-action

Create a release

1. Bump versions if needed (package.json, src-tauri/Cargo.toml, src-tauri/tauri.conf.json).
2. Commit to main.
3. Create and push a tag:
   • git tag v0.1.1
   • git push origin v0.1.1
4. Wait for the Release workflow to complete.
5. Verify assets in the GitHub Releases tab.

Unsigned build note

Because release builds are currently unsigned, users may need to run:

xattr -rd com.apple.quarantine /Applications/OpenSpeak.app

Optional: signing and notarization (recommended)

The workflow supports macOS signing/notarization if these repository secrets are set:

• APPLE_SIGNING_IDENTITY
• APPLE_CERTIFICATE (base64-encoded .p12)
• APPLE_CERTIFICATE_PASSWORD
• APPLE_ID
• APPLE_PASSWORD (app-specific password)
• APPLE_TEAM_ID

If these secrets are not set, builds can still complete, but distribution UX is typically better with signed/notarized artifacts.

License

Add your preferred license in LICENSE.