VoiceClutch

Private, free, and on-device voice dictation app for macOS with low-latency streaming transcription and LLM-powered corrections.

• Stay in flow - Dictate naturally with low-latency streaming and quick text injection into any active app.
• Get cleaner output - Optional LLM final pass improves punctuation, structure, and readability before insertion.
• Keep your data local - Speech recognition and transcript cleanup run on-device for privacy-first usage.

Features

Dictation Workflow & Controls

• Menu bar app - Runs as a lightweight macOS status bar utility with quick access to controls and preferences.
• Flexible interaction modes - Choose between Hold-to-talk and Press-to-talk based on your workflow.
• Configurable listening shortcut - Set a preferred modifier key or create a custom key combination.

Productivity Helpers

• Vocabulary manager - Add manual replacements, review learned corrections, and import/export vocabulary JSON.
• Clipboard recovery - Restores your previous clipboard contents after dictation injection.
• Media pause/resume - Optionally pauses active macOS media playback while dictating and resumes after.
• Microphone chimes - Optional start/stop sounds for listening state feedback.

Local Intelligence & Privacy

• On-device processing - Speech recognition and transcript post-processing run locally on your Mac.
• Optional LLM final pass - Enable local transcript cleanup and formatting refinement.
• Optional clipboard context - Allow clipboard-aware formatting context for the final pass.
• Privacy-first behavior - VoiceClutch does not collect analytics or usage telemetry.

Getting Started

1. Download the latest release, then move VoiceClutch.app to Applications.
2. Launch VoiceClutch and confirm the menu bar icon appears.
3. Grant requested permissions (microphone and accessibility) when prompted.
4. Place your cursor in any text field.
5. Press and hold Left Control to dictate.
6. Open Preferences to customize shortcut, interaction mode, and optional features.

Acknowledgements

• NVIDIA Nemotron Speech Streaming 0.6b CoreML by FluidInference - On-device low-latency streaming speech-to-text model.
• LFM2.5-1.2B-Instruct-MLX-4bit - Compact local large-langauge model for cleanup.
• mlx-swift-lm - Swift MLX language-model runtime.
• Apple Frameworks - Native frameworks for app core.

Feedback & Contributions

For issues, suggestions, or feature requests, please open a ticket.

License

MIT License