Super Voice Assistant

macOS voice assistant with global hotkeys - transcribe speech to text with offline models (WhisperKit or Parakeet) or cloud-based Gemini API, capture and transcribe screen recordings with visual context, and read selected text aloud with Gemini Live. Fast, accurate, and simple.

Demo

Parakeet transcription (fast and accurate):

parakeet.mov

Instant text-to-speech:

tts_compressed.mp4

Visual disambiguation for names:

visual-disambiguation-demo.mov

Features

Voice-to-Text Transcription

• Press Command+Option+Z for local offline transcription (WhisperKit or Parakeet)
• Press Command+Option+X for cloud transcription with Gemini API
• Choose your engine in Settings: WhisperKit models or Parakeet (faster, more accurate)
• Automatic text pasting at cursor position
• Transcription history with Command+Option+A

Streaming Text-to-Speech

• Press Command+Option+S to read selected text aloud using Gemini Live API
• Press Command+Option+S again while reading to cancel the operation
• Sequential streaming for smooth, natural speech with minimal latency
• Smart sentence splitting for optimal speech flow

Screen Recording & Video Transcription

• Press Command+Option+C to start/stop screen recording
• Automatic video transcription using Gemini 2.5 Flash API with visual context
• Better accuracy for programming terms, code, technical jargon, and ambiguous words
• Transcribed text automatically pastes at cursor position

Requirements

• macOS 14.0 or later
• Xcode 15+ or Xcode Command Line Tools (for Swift 5.9+)
• Gemini API key (for text-to-speech and video transcription)
• ffmpeg (for screen recording functionality)

System Permissions Setup

This app requires specific system permissions to function properly:

1. Microphone Access

The app will automatically request microphone permission on first launch. If denied, grant it manually:

• Go to System Settings > Privacy & Security > Microphone
• Enable access for Super Voice Assistant

2. Accessibility Access (Required for Global Hotkeys & Auto-Paste)

You must manually grant accessibility permissions for the app to:

• Monitor global keyboard shortcuts (Command+Option+Z/S/X/A/V/C, Escape)
• Automatically paste transcribed text at cursor position

To enable:

1. Go to System Settings > Privacy & Security > Accessibility
2. Click the lock icon to make changes (enter your password)
3. Click the + button to add an application
4. Navigate to the app location:
   • If running via swift run: Add Terminal or your terminal app (iTerm2, etc.)
   • If running the built binary directly: Add the SuperVoiceAssistant executable
5. Ensure the checkbox next to the app is checked

Important: Without accessibility access, the app cannot detect global hotkeys (Command+Option+Z/X/A/S/C/V, Escape) or paste text automatically.

3. Screen Recording Access (Required for Video Transcription)

The app requires screen recording permission to capture screen content:

• Go to System Settings > Privacy & Security > Screen Recording
• Enable access for Terminal (if running via swift run) or SuperVoiceAssistant

Installation & Running

# Clone the repository
git clone https://github.com/yourusername/super-voice-assistant.git
cd super-voice-assistant

# Install ffmpeg (required for screen recording)
brew install ffmpeg

# Set up environment (for TTS and video transcription)
cp .env.example .env
# Edit .env and add your GEMINI_API_KEY

# Build the app
swift build

# Run the main app
swift run SuperVoiceAssistant

The app will appear in your menu bar as a waveform icon.

Configuration

Text Replacements

You can configure automatic text replacements for transcriptions by editing config.json in the project root:

{
  "textReplacements": {
    "Cloud Code": "Claude Code",
    "cloud code": "claude code",
    "cloud.md": "CLAUDE.md"
  }
}

This is useful for correcting common speech-to-text misrecognitions, especially for proper nouns, brand names, or technical terms. Replacements are case-sensitive and applied to all transcriptions.

Usage

Voice-to-Text Transcription

Local (Cmd+Option+Z):

1. Launch the app - it appears in the menu bar
2. Open Settings to select and download a model (Parakeet or WhisperKit)
3. Press Command+Option+Z to start recording
4. Press Command+Option+Z again to stop and transcribe
5. Text automatically pastes at cursor
6. Press Escape to cancel

Cloud (Cmd+Option+X):

1. Set GEMINI_API_KEY in your .env file
2. Press Command+Option+X to start/stop recording
3. Text automatically pastes at cursor

Transcription engines:

• Parakeet v2: ~110x realtime, 1.69% WER, English - recommended for speed
• Parakeet v3: ~210x realtime, 1.8% WER, 25 languages
• WhisperKit: Various model sizes, good accuracy, more language options
• Gemini: Cloud-based, best for complex audio, requires internet

Text-to-Speech

1. Select any text in any application
2. Press Command+Option+S to read the selected text aloud
3. Press Command+Option+S again while reading to cancel the operation
4. The app uses Gemini Live API for natural, streaming speech synthesis
5. Configure audio devices via Settings for optimal playback

Screen Recording & Video Transcription

1. Press Command+Option+C to start screen recording
2. The menu bar shows "🎥 REC" while recording
3. Press Command+Option+C again to stop recording
4. The app automatically transcribes the video using Gemini 2.5 Flash
5. Visual context improves accuracy for code, technical terms, and homophones
6. Transcribed text pastes at your cursor position
7. Video file is automatically deleted after successful transcription

Note: Audio recording and screen recording are mutually exclusive - you cannot run both simultaneously.

When to use video vs audio:

• Video: Programming, code review, technical documentation, names, acronyms, specialized terminology
• Audio: General speech, quick notes, casual transcription

Keyboard Shortcuts

• Command+Option+Z: Start/stop audio recording and transcribe (WhisperKit - offline)
• Command+Option+X: Start/stop audio recording and transcribe (Gemini - cloud)
• Command+Option+S: Read selected text aloud / Cancel TTS playback
• Command+Option+C: Start/stop screen recording and transcribe
• Command+Option+A: Show transcription history window
• Command+Option+V: Paste last transcription at cursor
• Escape: Cancel audio recording (when recording is active)

Available Commands

# Run the main app
swift run SuperVoiceAssistant

# List all available WhisperKit models
swift run ListModels

# Test downloading a model (currently set to distil-whisper_distil-large-v3)
swift run TestDownload

# Validate downloaded models are complete
swift run ValidateModels

# Delete all downloaded models
swift run DeleteModels

# Delete a specific model
swift run DeleteModel <model-name>
# Example: swift run DeleteModel distil-large-v3

# Test transcription with a sample audio file
swift run TestTranscription

# Test live transcription with microphone input
swift run TestLiveTranscription

# Test streaming TTS functionality
swift run TestStreamingTTS

# Test audio collection for TTS
swift run TestAudioCollector

# Test sentence splitting for TTS
swift run TestSentenceSplitter

# Test screen recording (3-second capture)
swift run RecordScreen

# Test video transcription with Gemini API
swift run TranscribeVideo <path-to-video-file>
# Example: swift run TranscribeVideo ~/Desktop/recording.mp4

Project Structure

• Sources/ - Main app code
  • ModelStateManager.swift - Engine and model selection
  • AudioTranscriptionManager.swift - Audio recording and transcription routing
  • ScreenRecorder.swift - Screen recording with ffmpeg
• SharedSources/ - Shared components
  • ParakeetTranscriber.swift - FluidAudio Parakeet wrapper
  • GeminiStreamingPlayer.swift - Streaming TTS playback
  • GeminiAudioTranscriber.swift - Gemini API transcription
  • VideoTranscriber.swift - Gemini API video transcription
• tests/ - Test utilities
• tools/ - Model management utilities

License

See LICENSE for details.