How do you use Whisper for dictation on a Mac?

Whisper is an open source speech recognition model from OpenAI. You do not need to install it yourself to dictate on a Mac. Most people get the most from an app that runs a Whisper model and types into the app they are using. If you want the model itself, the open source route works, but it is a transcription tool first and needs a few setup steps.

Use an app for live dictation, and Whisper runs inside it

Several Mac apps run Whisper models on your machine, so you never touch the model directly. Handy is free and open source, works offline, and offers Whisper models in four sizes, Small, Medium, Turbo and Large ¹. Its README says you press a shortcut, speak, and the words appear in any text field ². Superwhisper's free plan uses local Whisper models without AI processing ³. VoiceInk credits whisper.cpp, the same engine described below, for running the Whisper model ⁴.

For recordings, Buzz describes itself as transcribing and translating audio offline on a personal computer ⁵.

Build whisper.cpp to run the model yourself

whisper.cpp is a C and C++ version of Whisper that runs well on Apple silicon. Its README calls Apple silicon a first-class target, optimized through ARM NEON, the Accelerate framework and Metal ⁶. The steps below follow the project's README, which we read for this guide.

1. Open Terminal, then download the project with git clone and move into the whisper.cpp folder.
2. Download a model file with the project's script, for example sh ./models/download-ggml-model.sh base.en. The English-only base.en model is small. Larger models need more memory.
3. Build the program with cmake -B build, then cmake --build build -j --config Release.
4. Convert your recording to 16-bit WAV. The README's example uses ffmpeg: ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav. The command-line example "currently runs only with 16-bit WAV files" ⁷, so this step is required.
5. Run the transcription with ./build/bin/whisper-cli -m models/ggml-base.en.bin -f output.wav.

The output appears in Terminal. That is a transcript of a file. It does not type into the app you are working in. For live dictation into other apps, use one of the apps above.

The Python package needs ffmpeg too

OpenAI's own repository installs Whisper with pip install -U openai-whisper. It also "requires the command-line tool ffmpeg to be installed on your system" ⁸, which you can install with Homebrew. The Python route is useful if you want to script transcription, but it is not a dictation tool.

Apple silicon makes the encoder faster

On Apple silicon, whisper.cpp can run the encoder on the Apple Neural Engine through Core ML. The project says that can be more than three times faster than CPU-only execution ⁹. Speed depends on your Mac and the model size you pick, so test the model on a real recording before you settle on one.

Accuracy and failure modes depend on the audio

Accuracy depends on the recording. In a 2025 test of one 7-minute-31-second episode, Whisper Large V3 Turbo had a word error rate of 1.5 percent ¹⁰. That is one clean recording. On the CHiME6 test set, Whisper Large V2 had a word error rate of 25.5 percent ¹¹, so the audio you use can change the result a lot.

Whisper also has a known failure mode. In a 2024 study of Whisper transcriptions, roughly 1 percent contained entire phrases or sentences that did not exist in the audio ¹². Read every transcript against the audio, especially for medical, legal or financial content.

What to do

• For live dictation, start with an app that runs Whisper for you, such as Handy or Superwhisper's free plan, and compare it with Apple Dictation on the same paragraph. Our question on free dictation apps for Mac lists the limits of each.
• For recordings, try Buzz, or the whisper.cpp steps above if you want full control.
• Build whisper.cpp only if you are comfortable in Terminal and want to script transcription. Our guide to running Whisper locally covers the trade-offs in more depth.
• Check every transcript against the audio. The glossary entry on word error rate explains how to measure the errors you find.

For the Apple comparison, see whether Apple Dictation is as good as Wispr Flow. For the overview of Mac dictation, see our best dictation app guide for Mac, and for privacy with Whisper, read our guide to Whisper privacy. For what "on-device" means in practice, see the glossary entry on on-device speech recognition.