GitHub - cydanix/whisperclip: Privacy-First Voice-to-Text with AI Enhancement for macOS

6 min read Original article ↗

✨ Features

🎤 Voice-to-Text Transcription

  • Two speech recognition engines:
    • Parakeet (default) - Fast and accurate using Apple Neural Engine, 25 European languages
    • WhisperKit - Multiple model sizes (216MB to 955MB) for different accuracy/speed trade-offs
  • Support for multiple languages with auto-detection
  • Real-time waveform visualization during recording
  • Audio file transcription - Import MP3, WAV, M4A, FLAC, and more

🤖 AI-Powered Text Enhancement

  • Local LLM processing for grammar correction and text improvement
  • Multiple AI models including Gemma, Llama, Qwen, and Mistral
  • Custom prompts for different use cases:
    • Grammar fixing and email formatting
    • Language translation
    • Custom text processing workflows

🔒 Privacy-First Design

  • 100% local processing - your voice never leaves your device
  • No cloud services, no data collection
  • Open source - audit the code yourself
  • Secure sandboxed environment

📝 AI Meeting Notes

  • Live meeting transcription with dual-channel audio (microphone + system audio)
  • Automatic speaker separation — "Me" vs "Others" with accurate timestamps
  • AI-generated summaries, action items, decisions, and follow-ups
  • Post-meeting Q&A — ask questions about any meeting and get AI-powered answers
  • Auto-detection of Zoom, Teams, Google Meet, Webex, Slack, Discord, and FaceTime
  • Configurable meeting hotkey (Control+M by default) to start/stop recording
  • Export meetings as Markdown or copy transcripts and summaries
  • Beautiful detail view with Summary, Transcript, Actions, and Q&A tabs

⚡ Productivity Features

  • Global hotkey support (Option+Space by default)
  • Hold to Talk mode - hold hotkey to record, release to stop
  • Auto-copy to clipboard
  • Auto-paste functionality
  • Auto-enter for instant message sending
  • Menu bar integration with background operation (runs without a visible window)
  • Start minimized option
  • Auto-stop recording after 10 minutes
  • Transcription history - Browse and search past transcriptions

🎨 User Experience

  • Beautiful dark-themed interface with modern sidebar navigation
  • Real-time recording visualization with animated effects
  • Recording overlay with customizable position
  • Comprehensive onboarding guide
  • Easy model management and downloads
  • Customizable shortcuts and prompts
  • Drag-and-drop audio file support

📋 Requirements

  • Apple Silicon Mac (M1, M2, M3, or later)
  • macOS 14.0 or later
  • 20GB free disk space (for AI models)
  • Microphone access permission
  • Accessibility permissions (for global hotkeys)
  • Apple Events permissions (for clipboard operations)

🚀 Installation

Download Pre-built App

  1. Visit whisperclip.com
  2. Download the latest release
  3. Drag WhisperClip.app to your Applications folder
  4. Follow the setup guide for permissions

Build from Source

# Clone the repository
git clone https://github.com/cydanix/whisperclip.git
cd whisperclip

# Build the app
./build.sh

# For development
./local_build.sh Debug
./local_run.sh Debug

🔧 Usage

Quick Start

  1. Launch WhisperClip from Applications or menu bar
  2. Grant permissions when prompted (microphone, accessibility)
  3. Download AI models through the setup guide
  4. Press Option+Space (or click Record) to start recording
  5. Press again to stop - text will be automatically copied to clipboard

Customization

  • Change hotkey: Settings → Hotkey preferences
  • Add custom prompts: Settings → Prompts → Add new prompt
  • Switch AI models: Setup Guide → Download different models
  • Configure auto-actions: Settings → Enable auto-paste/auto-enter

🤖 Supported AI Models

Speech-to-Text

Parakeet (default, recommended)

  • Fast transcription using Apple Neural Engine
  • 25 European languages supported
  • Optimized for Apple Silicon

WhisperKit (alternative)

  • OpenAI Whisper Small (216MB) - Fast, good quality
  • OpenAI Whisper Large v3 Turbo (632MB) - Best balance
  • Distil Whisper Large v3 Turbo (600MB) - Optimized speed
  • OpenAI Whisper Large v2 Turbo (955MB) - Maximum accuracy

Text Enhancement (Local LLMs)

  • Gemma 2 (2B/9B) - Google's efficient models
  • Llama 3/3.2 (3B/8B) - Meta's powerful models
  • Qwen 2.5/3 (1.5B-8B) - Alibaba's multilingual models
  • Mistral 7B - High-quality French company model
  • Phi 3.5 Mini - Microsoft's compact model
  • DeepSeek R1 - Advanced reasoning model

All models run locally using MLX for Apple Silicon optimization.

🔒 Privacy & Security

WhisperClip is designed with privacy as the cornerstone:

  • Local Processing Only: All voice recognition and AI processing happens on your device
  • No Network Requests: Except for downloading models from Hugging Face
  • No Analytics: No usage tracking, no telemetry, no data collection
  • Open Source: Full transparency - inspect the code yourself
  • Sandboxed: Runs in Apple's secure app sandbox
  • Encrypted Storage: AI models stored securely on device

🛠 Development

Project Structure

Sources/
├── WhisperClip.swift          # Main app entry point
├── ContentView.swift          # Main UI with sidebar navigation
├── MicrophoneView.swift       # Voice recording interface
├── FileTranscriptionView.swift # Audio file transcription
├── HistoryView.swift          # Transcription history browser
├── SharedViews.swift          # Shared UI components
├── AudioRecorder.swift        # Voice recording logic
├── VoiceToText*.swift         # Transcription engines (Parakeet, WhisperKit)
├── LLM*.swift                 # AI text enhancement
├── TranscriptionHistory.swift # History data management
├── ModelStorage.swift         # Model management
├── SettingsStore.swift        # User preferences
├── HotkeyManager.swift        # Global shortcuts
├── MeetingSession.swift       # Meeting lifecycle orchestration
├── MeetingRecorder.swift      # Dual-channel audio capture & transcription
├── MeetingAI.swift            # AI summaries, Q&A, and analysis
├── MeetingDetector.swift      # Meeting app auto-detection
├── MeetingStorage.swift       # Meeting notes persistence
├── MeetingModels.swift        # Meeting data models
├── MeetingNotesView.swift     # Meeting list & live recording UI
├── MeetingDetailView.swift    # Meeting detail with tabs
└── MeetingWaveformView.swift  # Audio waveform visualization

Dependencies

  • FluidAudio: Parakeet speech recognition with Apple Neural Engine
  • WhisperKit: Apple's optimized Whisper implementation
  • MLX: Apple Silicon ML framework
  • MLX-Swift-Examples: LLM implementations
  • Hub: Hugging Face model downloads

Building

# Debug build
./local_build.sh Debug

# Release build with code signing
./build.sh

# Notarization (requires Apple Developer account)
./notarize.sh

🤝 Contributing

We welcome contributions! Please see our contributing guidelines:

  1. Fork the repository
  2. Create a feature branch: git checkout -b feature/amazing-feature
  3. Make your changes and add tests
  4. Commit your changes: git commit -m 'Add amazing feature'
  5. Push to branch: git push origin feature/amazing-feature
  6. Open a Pull Request

Areas for Contribution

  • New AI model integrations
  • UI/UX improvements
  • Performance optimizations
  • Language support
  • Accessibility features
  • Documentation improvements

📄 License

WhisperClip is licensed under the MIT License - see the LICENSE file for details.

This means you can:

  • ✅ Use commercially
  • ✅ Modify and distribute
  • ✅ Use privately
  • ✅ Fork and create derivatives

Attribution required: Please include the original license notice.

🏢 About

WhisperClip is developed by Cydanix LLC.

🙏 Acknowledgments

  • Apple - WhisperKit and MLX frameworks
  • Senstella - FluidAudio and Parakeet models
  • OpenAI - Original Whisper models
  • Hugging Face - Model hosting and Hub library
  • ML Community - Open source AI models (Gemma, Llama, Qwen, etc.)

Made with ❤️ for privacy-conscious users

⭐ Star this repo if you find it useful!