WhisperTool
Free voice-to-text for Mac that lives in the menu bar. Hold ⌥ + ⌃, talk, let go — you see your words as you say them, and the text is typed straight into whatever app you're using.
This is a personal, non-commercial project — I built it to make my own day easier and I'm sharing it free for anyone who wants to try it. It started as a fork of WisprTool by FuturMinds — full credit to them for the original tool and idea.
Version 2: native Mac app (mac/)
Version 1 was Python + Tk and kept the Whisper model loaded on the GPU all day: 2.6 GB of memory even when you weren't talking. Version 2 is a single Swift file:
| v1 (Python) | v2 (Swift) |
|---|
| Memory while idle | 2,588 MB | 23 MB |
| CPU while idle | ~2% | 0% |
| Microphone | on all day | only while you hold the keys |
| Download | 174 MB | 1 MB |
- Menu bar app: no Dock icon, no window. Opens at login. Recent dictations one click away.
- Live text: while you hold the keys, a black island drops from the notch and shows what you're saying (Apple's on-device speech recognition).
- Whisper on Groq: with your own free Groq key, the audio goes to Whisper large-v3-turbo (~0.3 s per phrase, language detected per phrase), then an AI pass removes ums and self-corrections and fixes punctuation. Start with "this is an email…" and it formats an email.
- Nothing gets lost: if Groq fails, it pastes what your Mac heard on-device; if both fail, the audio is kept and the menu offers Retry the last audio.
- Never hangs on the mic: the audio engine is stopped on a background thread, so a slow CoreAudio teardown can't freeze the app (that was v1's worst bug).
Build it: cd mac && ./build.sh (Xcode command line tools; it signs ad-hoc if you don't have a developer certificate).
Privacy
- Without a key, transcription is Apple's on-device recognition: nothing leaves your Mac.
- With your Groq key, the audio of each dictation is sent to Groq to be transcribed, and the text to Groq to be tidied. Nothing goes to me.
- Your key lives only in a local
config.json (permissions 600), which is git-ignored.
Permissions
Microphone (to hear you), Speech Recognition (live text) and Accessibility (the global hotkey and pasting).
Mac app
Download WhisperTool.pkg from Releases. Requires Apple Silicon (M1 or later) and macOS 14 or later (v2.2; v2.1 needed macOS 26). Signed with Developer ID and notarized by Apple since v2.1: the installer puts it in Applications and opens it. (WhisperTool.dmg is still there for drag-and-drop.) To build it yourself: mac/build.sh, then sellar.sh pkg (see CHANGELOG).
Version 1 (Python, cross-platform)
The original Python app (main.py) is still here, for Windows or for fully local transcription with mlx/faster-whisper.
Install
git clone <your-fork-url> whispertool
cd whispertool
pip3 install -r requirements.txt
cp config.example.json config.json # optional: edit to enable AI polish
python3 main.py
For the optional AI polish, get a free Gemini API key at
aistudio.google.com/apikey and paste it into the
app's settings (or config.json). Never commit config.json — it's already git-ignored.
Credit
Original tool by FuturMinds
(YouTube). This fork adds the
Mac-specific GPU / overlay / AI changes listed above. Shared free, for anyone to use.
