The official Python package on macOS
- Install FFmpeg with Homebrew: brew install ffmpeg (the method the Whisper README gives for macOS).
- Install the package with pip install -U openai-whisper, in a Python version the README lists as expected to work (3.8–3.11).
- Use it on recordings, for example whisper memo.m4a --model turbo. It writes transcripts; it does not listen for a shortcut or type into Mail or Slack.
This is the most direct way to try Whisper on a Mac. It's a file tool, though, not dictation.
whisper.cpp on Apple Silicon: Metal and Core ML
whisper.cpp calls Apple Silicon a first-class platform, optimised with ARM NEON, the Accelerate framework, Metal and Core ML. Its README says that on Apple Silicon, inference runs fully on the GPU through Metal. It lists macOS on both Intel and Arm.
- Basic build: clone the repo, run sh ./models/download-ggml-model.sh base.en, then build with cmake -B build.
- Optional Core ML encoder on the Apple Neural Engine: install coremltools and the other listed Python packages, make sure Xcode and its command-line tools are installed (xcode-select --install), generate the model with ./models/generate-coreml-model.sh base.en, and rebuild with -DWHISPER_COREML=1.
- Expect the first Core ML run to be slow. The README explains that the Neural Engine service compiles the model for your device on first use.
- Input must be 16-bit WAV for whisper-cli. Live microphone input via whisper-stream needs SDL2 and prints text in Terminal.
Mac apps that wrap a speech model
If you'd rather not build anything, Mac apps package recording, a model and text output. Check each app's own pages for which models it runs and whether processing stays on your Mac.
- MacWhisper: centred on transcription. Its site covers files, meetings and app audio, says it uses local models for file transcription, and also offers dictation.
- Superwhisper: a dictation app for macOS (and Windows). Its site says it works offline and pastes into the app you're using.
- Whisper Key Local: open source, runs on macOS as well as Windows. Its default Mac shortcut is Fn+Ctrl to start and Fn to stop and transcribe.
Mac permissions any dictation tool will ask for
To hear you, a Mac dictation app needs Microphone access. To put text into other apps or watch for a global shortcut, it usually needs Accessibility, or a related input permission, under System Settings → Privacy & Security. A whisper.cpp script run in Terminal asks for these permissions on Terminal's behalf. Whispek asks for Microphone, Accessibility and related permissions when a feature first needs them.
Whisper on a Mac vs Whispek for Mac
Whispek for Mac doesn't use Whisper. Speech is recognized in the cloud, so there's no model to download and no Metal or Core ML build to maintain. You just need to be online.
| Whisper on your Mac | Whispek for Mac (beta) | |
|---|---|---|
| Install | Homebrew + Python package, or a whisper.cpp build + model files; optional Xcode tools for Core ML | Download the Apple Silicon or Intel .dmg (version 0.1.16 beta), drag to Applications. Requires macOS 14.2 or later |
| First open | — | The beta is not notarized, so macOS may block the first launch. Right-click → Open, or allow it in System Settings → Privacy & Security |
| Dictate into apps | Needs a wrapper app or script | Press Right Option (default; change it on the Home screen), speak, review. If the text can't go into the app, copy and paste it |
| Writing tools | Depends on the app | Auto cleanup, Dictionary, Transforms, Snippets, History, Whispek Assistant |
| Offline | Local models can work offline | Internet required; History stays on your Mac |
| Cost | Whisper is MIT-licensed; apps have their own pricing | Free with an allowance every 7 days; Pro $12/month or $99/year ($8.25/month equivalent, billed yearly), before tax; all current Pro features unlimited |
Mac is in beta. Found a bug? Tell us from the app or via Contact.
Which Mac route fits you
- Transcribing interviews, lectures or meetings on-device → MacWhisper or the Whisper CLI/whisper.cpp.
- Tinkering, scripting or comparing models on Apple Silicon → whisper.cpp with Metal, adding Core ML if you like.
- Writing emails, messages and AI prompts by voice with nothing to set up → Whispek for Mac. Hold Right Option and talk. Free every 7 days.
- Dictating prompts for Claude Code or Codex in Terminal → with Whispek, dictate, review, then copy and paste into Terminal.