Voice to Text — Local Whisper Transcription for VS Code

Privacy-first voice-to-text for macOS. Record, transcribe, and paste — all locally with OpenAI Whisper. No cloud, no API keys, no data leaves your machine.

Features

• 100% local transcription — powered by whisper.cpp with Metal GPU acceleration on Apple Silicon
• Four models to choose from:
  • Large v3 Turbo — Best accuracy, multilingual + translation (~1.5 GB)
  • Small (Multilingual) — Good balance, supports translation (~466 MB)
  • Small English — Fast with good accuracy, English only (~466 MB) ← default
  • Base English — Fastest, English only (~142 MB)
• One-key recording — Press ⌘⇧; (Command + Shift + semicolon) to toggle
• Live audio waveform — Record button shows real-time voice visualization while recording
• Streaming preview — See partial transcription in the status bar while still speaking
• Smart punctuation — Auto-capitalizes sentences, fixes "i" → "I", adds trailing periods
• System-wide paste — Transcribed text is pasted at your cursor, wherever it is
• Transcription history — Browse, search, copy, and manage past transcriptions in the sidebar
• Inline settings — Switch models, language, and mic directly from the history panel
• Model pre-warming — Whisper loads into memory at startup for faster first transcription
• Audio preprocessing — Noise reduction, volume normalization, and compression for better accuracy
• Cancel with Escape — Discard a recording without transcribing
• Zero configuration — First-time setup handles everything automatically

Prerequisites

Before installing, make sure you have:

1. macOS on Apple Silicon (M1/M2/M3/M4) — Intel Macs work too, but without Metal GPU acceleration
2. Xcode Command Line Tools — open Terminal and run:

   xcode-select --install
   

3. Homebrew — if you don't have it, install from brew.sh:

   /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
   

That's it. The extension handles everything else (ffmpeg, cmake, whisper.cpp, model downloads) automatically on first use.

Getting Started

First Run

1. Press ⌘⇧; or click the 🎙 button in the Voice to Text sidebar
2. The extension will detect that first-time setup is needed and ask to run it
3. Click Run Setup — this will:
   • Install ffmpeg and cmake via Homebrew (if not already installed)
   • Clone and build whisper-cli from whisper.cpp v1.7.5 with Metal GPU support
   • Download the Small English model (~466 MB) from Hugging Face
4. Setup takes 2–5 minutes depending on your internet speed. You'll see progress notifications.

Recording

1. Press ⌘⇧; to start recording — the button shows a live audio waveform
2. Speak naturally
3. Press ⌘⇧; again to stop — the audio is transcribed locally and pasted at your cursor
4. Press Escape to cancel a recording without transcribing

History Panel

Open the Voice to Text sidebar (mic icon in the activity bar) to see:

• All past transcriptions with timestamps and duration
• Start Recording button with live waveform visualization
• Copy / Delete buttons on each entry
• Search box to filter past transcriptions
• Settings section to change model, language, and microphone

Switching Models

From the history panel's ⚙ Settings section or the Command Palette (⌘⇧P → "Voice to Text: Switch Whisper Model").

Multilingual & Translation

Select a non-English language in Settings. If you're on an English-only model, you'll be prompted to switch to a multilingual one. Non-English speech is automatically translated to English using whisper.cpp's built-in translation.

Settings Reference

How It Works

1. Recording — ffmpeg captures audio from your microphone via macOS AVFoundation at 16 kHz mono with noise reduction and volume normalization
2. Transcription — whisper-cli processes the audio locally using the selected Whisper model with Metal GPU acceleration
3. Paste — Text is inserted directly via the VS Code editor API, or pasted via clipboard + AppleScript for non-editor inputs

All processing happens on your machine. No audio or text is sent anywhere.

Troubleshooting

"Setup is still running..." — The first-time setup is in progress. Wait for it to complete.

No speech detected — Check that the correct microphone is selected in Settings. Device 0 is often a virtual device that captures silence.

Want to reset everything?

rm -rf ~/.voicetotext

Data Storage

~/.voicetotext/
├── whisper.cpp/          # whisper.cpp source + built binary
├── models/               # Downloaded Whisper model files
└── history.json          # Transcription history

Temporary recordings are stored in your system temp directory and deleted after transcription.

License

MIT