Skip to content

About

On-device macOS menu bar app for speech-to-text using WhisperKit — no cloud, no API keys, all local.

Topics

Resources

Contributing

Stars

2 stars

Watchers

0 watching

Forks

Repository files navigation

VoiceInput

A macOS menu bar app that converts speech to text using on-device AI. Powered by WhisperKit, all processing happens locally on your Mac — no cloud, no API keys, no data leaves your device.

Features

  • On-device processing — All transcription runs locally via WhisperKit + Core ML. Zero network calls.
  • Multi-language support — 7 selectable transcription languages: English, Korean, Japanese, Chinese, Spanish, French, and German, plus auto-detect.
  • Last transcription shortcut — Use a configurable shortcut to copy and paste your most recent transcription again.
  • Global hotkey — Start/stop recording from any app without switching windows.
  • Smart text insertion — Automatically pastes transcribed text into the active app with Cmd+V, with Accessibility API and keyboard simulation as fallbacks.
  • Multiple Whisper models — Choose the model that fits your speed/accuracy needs.
  • Menu bar app — Lives in the menu bar, stays out of your way.
  • Apple Silicon optimized — Runs on Core ML with Apple Neural Engine acceleration.

Requirements

  • macOS 14.0 (Sonoma) or later
  • Apple Silicon (M1/M2/M3/M4)

Installation

Build from Source

# Clone the repository
git clone https://github.com/junghyunJJ/voiceinput.git
cd voiceinput

# Install locally and launch as a macOS app
./Scripts/build-dev.sh

This creates ~/Applications/VoiceInput.app, signs it for local development, and launches it. Use swift build only when you want to check that the package compiles without creating or launching the app bundle.

For a signed distribution build with notarization and DMG packaging:

./Scripts/build-release.sh

Usage

  1. Launch — VoiceInput appears as an icon in the menu bar.
  2. Grant permissions — Allow Microphone and Accessibility access when prompted.
  3. Download your first model — Open Settings > Model and click Download on the model you want.
  4. Activate model — After download, click Use for that model.
  5. Record — Click the menu bar icon or use the global hotkey to start recording. Speak, then stop recording — the transcribed text is automatically inserted at your cursor.

Note: Internet is required only for the first model download.

Shortcuts

  • Recording: Option + Space by default (global)
  • Copy Last Transcription: Configurable in Settings (global)

Recording Modes

  • Toggle — Press once to start recording, press again to stop.
  • Push-to-Talk — Hold the shortcut to record, release to stop.

Change Shortcuts / Mode

  1. Open menu bar icon > Settings... > General
  2. In Hotkey, click a shortcut field and press your new key combination
  3. Choose Toggle or Push-to-Talk

Model Download Progress

  • While downloading, progress is shown in Settings > Model and also in the menu bar panel.
  • The app shows both progress bar and percentage.

Available Models

Model Size Speed Accuracy Best for
tiny ~75 MB Fastest Basic Quick notes, simple phrases
base ~145 MB Fast Good Everyday use
small ~465 MB Moderate Better General purpose (default)
medium ~1.5 GB Slower High Longer dictation with better accuracy
large-v3 ~3.0 GB Slowest Best Maximum accuracy
large-v3-turbo ~632 MB Faster than large-v3 High Better speed while keeping strong accuracy

For the highest transcription accuracy, we recommend large-v3.

Permissions

VoiceInput requires two macOS permissions:

  • Microphone — To capture audio for transcription.
  • Accessibility — To type transcribed text into the active application.

Both permissions are requested on first launch and can be managed in System Settings > Privacy & Security.

With Auto-insert text at cursor enabled, transcription uses a normal Cmd+V paste so browser editors receive input events. VoiceInput restores the previous clipboard after the paste, unless you copied something new in the meantime. If Accessibility access is unavailable, the transcription stays on the clipboard for manual paste.

Models and tokenizers are stored in ~/Library/Application Support/Voice Input/Models/, so model loading does not require access to your Documents folder. Caches from older versions in ~/Documents/huggingface/ are left intact; copy their models folder into the new location to reuse downloads, or download the model again in Settings.

Project Structure

VoiceInput/
├── Package.swift                 # Swift Package Manager manifest
├── Package.resolved              # Dependency lock file
├── Scripts/
│   ├── build-dev.sh              # Development build & launch
│   └── build-release.sh          # Release build, sign, notarize, DMG
├── VoiceInput/
│   ├── App/
│   │   ├── VoiceInputApp.swift       # App entry point
│   │   └── AppDelegate.swift         # NSApplicationDelegate
│   ├── Models/
│   │   ├── AppSettings.swift         # User preferences
│   │   └── RecordingState.swift      # Recording state machine
│   ├── Services/
│   │   ├── Audio/
│   │   │   ├── AudioService.swift    # Microphone capture
│   │   │   └── AudioBuffer.swift     # Audio sample buffer
│   │   ├── TextInsertion/
│   │   │   ├── TextInsertionManager.swift   # Insertion strategy coordinator
│   │   │   ├── AccessibilityInserter.swift  # AX API insertion
│   │   │   ├── KeyboardSimulator.swift      # CGEvent key simulation
│   │   │   └── ClipboardInserter.swift      # Clipboard fallback
│   │   ├── Transcription/
│   │   │   ├── TranscriptionEngine.swift    # Engine protocol
│   │   │   ├── WhisperKitEngine.swift       # WhisperKit implementation
│   │   │   └── TranscriptionResult.swift    # Result model
│   │   ├── HotkeyManager.swift       # Global hotkey registration
│   │   ├── ModelManager.swift         # Model download & management
│   │   └── PermissionsManager.swift   # Permission checks
│   ├── ViewModels/
│   │   └── AppViewModel.swift        # Main view model
│   ├── Views/
│   │   ├── MenuBarView.swift         # Menu bar UI
│   │   ├── SettingsView.swift        # Settings window
│   │   ├── OnboardingView.swift      # First-launch setup
│   │   └── RecordingOverlayView.swift # Recording indicator overlay
│   ├── Utilities/
│   │   └── Constants.swift           # App-wide constants
│   └── Resources/
│       ├── Info.plist                # App metadata
│       └── VoiceInput.entitlements   # Sandbox entitlements
└── VoiceInputTests/
    ├── AudioBufferTests.swift
    ├── RecordingStateTests.swift
    └── TranscriptionResultTests.swift

Tech Stack

  • Language: Swift 5.9
  • UI: SwiftUI + AppKit (menu bar)
  • Speech-to-Text: WhisperKit (Core ML)
  • Build System: Swift Package Manager
  • Minimum Target: macOS 14.0

Contributing

See CONTRIBUTING.md for guidelines on how to contribute.

License

This project is licensed under the MIT License — see the LICENSE file for details.

Acknowledgments

  • WhisperKit by Argmax — On-device speech recognition for Apple Silicon
  • Whisper by OpenAI — The model architecture behind the transcription

About

On-device macOS menu bar app for speech-to-text using WhisperKit — no cloud, no API keys, all local.

Topics

Resources

Contributing

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages