Skip to content

ConstantinSchreiber/mumbletype

Repository files navigation

Mumbletype

macOS menubar app for global voice-to-text input powered by OpenAI's state-of-the-art transcription models. Press a hotkey anywhere, speak, and your words are typed at the cursor — accurately.

Setup

Requires macOS and an OpenAI API key.

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Set your API key in Preferences after launching (or create ~/Library/Application Support/Mumbletype/.env containing OPENAI_API_KEY=sk-...).

Usage

source venv/bin/activate
python mumbletype.py
  • ⌃D (configurable) — Hit once to record, hit again to transcribe and paste
  • Click the menubar mic icon for model selection, usage stats, Start at Login, and preferences
  • Change the hotkey in Preferences → Record Hotkey → Change…

The hotkey is registered system-wide via Carbon RegisterEventHotKey: it is consumed by Mumbletype and never reaches the focused app, and no Accessibility or Input Monitoring permission is needed for listening. On first run macOS will still prompt for Microphone access (recording) and Accessibility (only for the synthetic ⌘V paste).

Start at Login (menubar toggle) installs a LaunchAgent that also restarts the app automatically if it ever crashes. Quitting from the menu stays quit.

Models

Model Cost Notes
GPT-4o Mini Transcribe $0.003/min Default, fast and cheap
GPT-4o Transcribe $0.006/min Higher accuracy
Whisper-1 $0.006/min Original Whisper model

Switch models from the menubar dropdown or Preferences window.

How it works

Mumbletype runs as a menubar-only app (no dock icon). The global hotkey toggles recording with sounddevice; audio is sent in-memory to the OpenAI transcription API and the result is pasted at the cursor via the clipboard + a synthetic ⌘V. Your previous clipboard contents — including images and rich text — are restored afterwards (unless you copied something in the meantime, in which case your copy wins). A floating waveform pill appears bottom-center of the screen your cursor is on during recording and transcription, and flashes red if transcription fails.

Configuration lives in ~/Library/Application Support/Mumbletype/; logs in ~/Library/Logs/Mumbletype.log.

License

MIT

About

macOS menubar app for global voice-to-text input powered by OpenAI's state-of-the-art transcription models. Press a hotkey anywhere, speak, and your words are typed at the cursor accurately.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages