How to Use Whisper on Mac Without Coding (Apps and Terminal)

The easiest way to use Whisper on a Mac without coding is a Whisper Mac app that bundles the model. You install the app, download a model inside it, and then click record or drop in a file. Which app fits depends on the job:

• MacWhisper transcribes audio and video files.
• Superwhisper dictates into any app.
• Gubb turns short English voice notes into text in a notes window.

If you're comfortable typing a few commands, the free whisper.cpp project runs Whisper in Terminal with the full choice of models and languages.

We make Gubb, so weigh our view accordingly. We've tried to be fair to every option here, including the free ones.

If you want to know what Whisper is before installing anything, read what OpenAI Whisper is and how the free, local and API versions differ.

Whisper Mac apps and other ways to run it

Whisper is OpenAI's open-source speech recognition model. It isn't built into macOS. Apple Dictation uses Apple's own speech engine.

If all you need is to talk into any text field, Apple Dictation may be enough. It's free, and on Apple silicon Macs it processes supported languages on your device. On Intel Macs, or in languages without on-device support, your speech is sent to Apple. See how to use Dictation on Mac.

You need Whisper itself when you want offline transcription of recordings, a specific open model, or one of the apps built around it. Here are the main routes.

Pick MacWhisper if you have recordings such as interviews, lectures or meetings and want a transcript or subtitle file. Version 15 needs macOS 15 or later, and batch transcription and speaker recognition need Pro. The App Store version is called "Whisper Transcription" and is sold separately from the direct version.

Pick Superwhisper if you want speech to replace typing in every app, or you need languages other than English. It needs macOS 14 or later. Superwhisper says its offline models only run really well on Apple silicon, and that Intel Macs do better with its cloud models.

Pick Gubb if you mostly dictate quick English thoughts and want them in a scratchpad that also has to-dos, clipboard history and timers, for a one-time price.

Pick whisper.cpp if you don't mind a few Terminal commands and want it free, with full control over models and languages.

For a wider comparison, see the best dictation apps for Mac and Superwhisper vs MacWhisper.

Which model size to choose

The main Whisper sizes are tiny, base, small, medium and large. Larger models make fewer mistakes but transcribe more slowly and need more memory and disk space. Some sizes also come in English-only versions, whose names end in .en.

Gubb offers three sizes and leaves out Tiny and Base on purpose.

In any app, start with the small model and move up a size only if you keep correcting words. The bigger files take real disk space, so delete the ones you stop using.

Set up Whisper in Gubb

Gubb runs Whisper locally through whisper.cpp. Once a model is downloaded and set as the active model, recordings are transcribed on your Mac without an internet connection.

Before you start

• Download Gubb from gubb.net/download (7-day free trial) or get it from the Mac App Store.
• Check that you have free disk space for the model you want (see the table above).
• macOS asks for microphone access the first time you record. Allow it.

Download and select a model

1. Open Gubb and move the pointer to the right edge of the note. A column of buttons appears.
2. Click Settings (the gear).
3. Choose Voice to Text in the settings sidebar.
4. Make sure the Voice to Text switch is on.
5. Under Transcription Provider, choose Local Models. This is the default.
6. Under Available Models, click Download next to Whisper Small, or the size you picked.
7. Wait for the "Download Complete" message. The progress bar may not move during the download, so don't give up because it looks stuck.
8. Under Active Model, select the model you just downloaded. It shows an Active badge.
9. Click outside the Settings panel, or click the close (X) button.
   Don't skip step 8. Without an active model, Gubb falls back to an OpenAI request with no API key, which fails with "Transcription Error". Restarting Gubb also activates a model, but only the first downloaded one in the order Small, Medium, Large.

Record and transcribe

1. Click in the note where you want the text to go.
2. Move the pointer to the right edge and click Speech to Text (the microphone icon). A pill appears at the bottom of the note reading "Click to start recording (Local Model)".
3. Click the microphone to start, and speak.
4. Click Stop. The pill shows "Processing audio...", then the text appears at your cursor, followed by a space.
   Three habits save you trouble:

• Keep each recording under 30 seconds. Gubb's local engine transcribes only the first 30 seconds, so record longer thoughts in several takes.
• Record for at least 1.5 seconds. If you press Stop sooner, Gubb shows "Keep recording".
• Speak English. Local transcription in Gubb is fixed to English, and there's no language setting.

What Gubb forgets between launches

Gubb doesn't save your Active Model choice. At each launch it activates the first downloaded model in the order Small, Medium, Large.

So if you have both Small and Large downloaded, Gubb starts with Small every time. To use Large, select it again after each launch, or delete Small under Available Models.

Models are stored in Gubb's sandbox container. For the direct download that's ~/Library/Containers/com.gubbapp.direct/Data/Library/Application Support/gubb/models/whisper/; the App Store version uses com.gubbapp.release in place of com.gubbapp.direct. Clicking Delete next to a model in Settings removes its file.

Gubb is a $59 one-time purchase, and you can try the direct download free for 7 days.

Run Whisper from Terminal with whisper.cpp

This route is free and has none of the app limits above: any model size, any language Whisper supports, any recording length. In return you type a few commands, and you transcribe audio files rather than dictating live.

If a command below has changed, the whisper.cpp repository README has the current version.

Install the tools

Install Apple's Command Line Tools, which include git and a compiler:

xcode-select --install

Install CMake (to build the project) and ffmpeg (to convert audio). This needs Homebrew:

brew install cmake ffmpeg

To remove them later, run brew uninstall cmake ffmpeg.

Build whisper.cpp and download a model

Download the source code into a whisper.cpp folder in your current directory, then move into it:

git clone https://github.com/ggml-org/whisper.cpp.git
cd whisper.cpp

Download the English-only base model into the models folder. Swap base.en for small, medium and so on to get other sizes:

sh ./models/download-ggml-model.sh base.en

Compile the program. The first build takes a while:

cmake -B build
cmake --build build -j --config Release

Gubb's local models are the same kind of ggml files, downloaded from the whisper.cpp model repository on Hugging Face.

Transcribe a recording

Convert your recording to the 16 kHz mono WAV format whisper.cpp expects:

ffmpeg -i recording.m4a -ar 16000 -ac 1 -c:a pcm_s16le recording.wav

Transcribe the WAV file with the model you downloaded. The transcript prints in Terminal, and -otxt also saves it as recording.wav.txt in the same folder:

./build/bin/whisper-cli -m models/ggml-base.en.bin -f recording.wav -otxt

whisper-cli assumes English unless told otherwise. For other languages, download a multilingual model (one without .en, such as small), point -m at it, and add -l auto to detect the language or a code such as -l de.

None of these commands change macOS settings. To remove whisper.cpp and its models, delete the whisper.cpp folder.

Limits to expect

Gubb

• English only, and only the first 30 seconds of each recording.
• Text goes into the open Gubb note only. It doesn't type into other apps, and there's no global hotkey to start recording.
• It doesn't transcribe audio or video files. There's no microphone picker; it records from your Mac's default input.
• Settings also list an OpenAI API provider, but as of version 1.0.1 it shows "OpenAI API is not configured" even after you save a key. Use Local Models.

MacWhisper

• Mac only, and recent versions need macOS 15 or later.
• Batch transcription, speaker recognition and the AI service integrations need Pro.

Superwhisper

• Paid tiers are a subscription or a $249.99 lifetime license (third-party figures, as of November 2026).
• Offline models run best on Apple silicon.

whisper.cpp

• No app window. You work with files and commands.
• whisper-cli doesn't record from your microphone. Record first, then transcribe.

If it doesn't work

"Recording Failed" in Gubb. Gubb couldn't start recording. Open System Settings, search for "Microphone", and turn on Gubb in the list under Privacy & Security. Also check that a model is downloaded under Available Models, then try again.

"Transcription Error" / "Failed to transcribe audio" in Gubb. Usually no model is active, often right after a download. Open Settings ▸ Voice to Text and select the model under Active Model. In this state Gubb sends the clip to OpenAI with no API key, so the request fails and the audio leaves your Mac. Select the model before you record. A clip that's too short or silent can also cause this error.

"Voice to Text Disabled". The Voice to Text switch in Settings is off. Turn it on.

The transcript stops mid-sentence. You went past 30 seconds. Record in shorter takes.

"Download Failed". The model download from Hugging Face didn't finish, most often because of the network. Check your connection and free disk space (3000 MB for Large), then click Download again.

whisper.cpp: command not found: brew or cmake. Install Homebrew from brew.sh, then run brew install cmake ffmpeg again.

whisper.cpp turns non-English speech into English or nonsense. You used a .en model. Download a multilingual model and add -l auto.

FAQ

Do I need the internet to use Whisper on a Mac?
Only to download the app and the model. After that, transcription runs on your Mac in Gubb (with an active local model), whisper.cpp and the offline modes of MacWhisper and Superwhisper. Gubb's direct download also checks its license online about once a week and asks you to reconnect after 30 days without a check.

Is Whisper free?
The model and whisper.cpp are free. The apps charge for some features, and OpenAI's hosted Whisper API bills per use. See what the OpenAI Whisper API costs.

Can Whisper type into any app on my Mac?
Not by itself. Superwhisper does this; Gubb inserts text only into its own notes. For free typing in any app, use Apple Dictation, which isn't Whisper.

Which Whisper model should I start with?
The small model. Move to Medium only if you keep fixing words and your Mac has 8 GB of RAM or more.