Whisper for Mac, running locally and typing into any app
By Ryleigh Newman · Published · Updated
Whisper for Mac, run locally, is what YapToText is: a free Mac app that runs OpenAI's Whisper Large v3 Turbo (Q5) on your Mac and types what you say into any app. It runs Whisper through whisper.cpp, with no Python, no Terminal and no account. The model ships inside the App Store download, and your voice never leaves your Mac.
I built it because my hands make typing difficult and I needed dictation that works in every app without sending my voice anywhere. This page covers what Whisper is, the exact model file the app uses, what changes when you run Whisper yourself, and where the code lives.
Download for free. No account, nothing locked, macOS 14 or later on Apple silicon. Source on GitHub.
On this page: 11 sections
Steps
- Install YapToText from the Mac App Store; Whisper Large v3 Turbo is inside.
- Allow the microphone, then turn on Accessibility for YapToText in System Settings.
- Click where you want the words in any app.
- Tap Right Command, talk, and tap it again.
What Whisper is
Whisper is OpenAI's speech recognition model. OpenAI calls it a general-purpose speech recognition model trained on a large dataset of diverse audio, and it released the code and the model weights under the MIT License. That license is why a free app can ship it.
Whisper Large v3 Turbo is the model YapToText runs. OpenAI describes turbo as an optimized version of large-v3 that offers faster transcription with a minimal degradation in accuracy, and its model card on Hugging Face says the decoding layers were cut from 32 to 4.
whisper.cpp is what makes it practical on a Mac: an MIT-licensed C and C++ implementation of Whisper whose README calls Apple silicon a first-class citizen, optimized through ARM NEON, Accelerate, Metal and Core ML. Whisper is the model, whisper.cpp is the engine, and YapToText is the app that types the result where your cursor is.
The Whisper model inside YapToText
- The file. ggml-large-v3-turbo-q5_0.bin, 574 MB, from the ggerganov/whisper.cpp repository on Hugging Face. Its SHA-256 matches the published file, so it is the standard 5-bit build, which I have not fine-tuned, retrained or modified.
- Why the 5-bit build. I picked it because, on my own dictations, it heard almost exactly what the full model heard, with a third of the memory. It became the default in 1.3.1, whose release notes say dictation got over twice as fast and the app about a gigabyte smaller.
- The engine. A copy of whisper.cpp built into the app with Metal and Accelerate, with llama.cpp beside it for the cleanup model.
- The licenses. The model is MIT, and the app is GPL-3.0.
What I did tune is everything around the model. Background noise is reduced and quiet speech is lifted before Whisper hears it, the half second before you press the key is kept so first words are not clipped, noisy audio gets deeper decoding, and hallucinated speaker labels and stage directions are removed from what comes back. Your dictionaries also prime Whisper with your own words. The model is stock; the pipeline around it is where the work went.
Pick another Whisper model
The AI Models page has a Dictation Model Library with ten more Whisper models, each downloaded once from Hugging Face and run on your Mac. The app's own rule of thumb: bigger models hear better, and smaller ones are lighter on battery and memory.
| Model | Size the app shows | Languages |
|---|---|---|
| Whisper Tiny (English) | 75 MB | English only |
| Whisper Base (English) | 142 MB | English only |
| Whisper Small (English) | 466 MB | English only |
| Whisper Base | 142 MB | Multilingual |
| Whisper Small | 466 MB | Multilingual |
| Whisper Medium | 1.5 GB | Multilingual |
| Whisper Large v3 Turbo (Q5), bundled and the default | 574 MB | Multilingual |
| Whisper Large v3 Turbo (Q8) | 834 MB | Multilingual |
| Whisper Large v3 Turbo | 1.5 GB | Multilingual |
| Whisper Large v3 (Q5) | 1.1 GB | Multilingual |
| Whisper Large v3 | 3.0 GB | Multilingual |
Under Your own models you can add any whisper.cpp GGML .bin file; the app recognizes it from the file itself and rejects anything else. Every model carries star ratings for accuracy and speed, and on the Energy page, Switch models with the power source can run a lighter model on battery. The AI models guide and the energy guide have the details.
Run Whisper yourself, or use an app
You can run Whisper on a Mac without any app, and it is a good way to learn how it works. Here is what each route asks of you.
Checked October 3, 2026, on each project's own README.
| OpenAI's whisper | whisper.cpp | YapToText | |
|---|---|---|---|
| Install | pip install -U openai-whisper, with Python 3.8 to 3.11 and ffmpeg | Clone the repository, download a model with its script, build with CMake | The Mac App Store, with the model inside |
| How you use it | A Terminal command or Python, on audio files | whisper-cli on 16-bit WAV files; whisper-stream for a live microphone, with SDL2 | A key in any app, and audio files on the Utility page |
| Models | tiny to large, plus turbo | Whisper models in ggml format, including quantized ones | Large v3 Turbo (Q5) inside, ten more in the library, or your own ggml file |
| Macs | Not stated | Intel and Arm | Apple silicon, macOS 14 or later |
| Price and license | Free, MIT | Free, MIT | Free, GPL-3.0 |
Running it yourself wins on control: any model, any flag, and Intel Macs too, since whisper.cpp lists Intel and Arm Macs as supported. YapToText wins on everything after the transcript: it types into the app you are in, cleans up with a local model if you want, keeps a History, and needs no setup. The only command to remember is Right Command. Run Whisper yourself to learn and tinker; use an app when you just want the words.
To run it yourself, the local Whisper guide has the whisper.cpp commands and what each model costs in disk space and memory.
For recordings rather than live dictation, the audio file transcription page covers the Utility page, and the MacWhisper alternative page compares YapToText with MacWhisper. The comparison page sets superwhisper, Wispr Flow, MacWhisper and VoiceInk side by side.
Get local Whisper running in three steps
Step 1: Install it
- You: Get YapToText from the Mac App Store. The download is about 2.9 GB, because both models are inside.
- YapToText: Arrives ready to dictate offline, with nothing else to download.
- Check: The AI Models page lists Whisper Large v3 Turbo (Q5) under Your defaults.
Step 2: Grant two permissions
- You: Allow the microphone when macOS asks, then press Grant next to Accessibility on the Home page and switch YapToText on in System Settings, Privacy & Security, Accessibility.
- macOS: Asks for the microphone once. Accessibility is a switch only you can turn on.
- Check: The Get set up card disappears from the Home page, which happens once both are on.
Step 3: Dictate
- You: Click where the words go, tap Right Command, talk, and tap it again.
- YapToText: Runs Whisper on your Mac and pastes the result at your cursor.
- Check: The words are there. The first dictation after launch can take a moment while the model loads.
Homebrew works too, with one extra step. The Homebrew build has no models inside, so download a speech model before your first dictation.
Terminal: Install YapToText from my Homebrew tap.
brew install --cask ryleighnewman/yaptotext/yaptotextWhat happens: YapToText lands in your Applications folder after a download of about 12 MB. It cannot dictate until you download a speech model on the AI Models page. Homebrew is on 1.5.2, and the App Store has 1.5.3.
The install page has both routes, and the first dictation guide walks through the panel and keys.
The open-source repository
The code is on GitHub at github.com/ryleighnewman/YapToText, under GPL-3.0. You can read exactly how audio reaches Whisper and what happens to the text afterward, open an issue, or build it yourself with Xcode 26 or later. Every line is public, so you never have to take my word for where your voice goes.
Terminal: Get the source and open it in Xcode.
git clone https://github.com/ryleighnewman/YapToText.git
open YapToText/YapToText.xcodeprojWhat happens: Xcode opens the project. Choose the YapToText scheme and press Command R. The app builds sandboxed, the way it ships, but without the models, so download a speech model on the AI Models page before your first dictation.
One honest gap: GitHub and Homebrew are on 1.5.2, and the App Store has 1.5.3.
If Whisper mishears you
Climb this ladder, smallest step first:
- Teach it your words. Add names and terms on the Dictionaries page; they prime Whisper, not just fix its output. The dictionaries guide shows how.
- Set your language. The Language picker on the AI Models page starts on your Mac's system language. The languages page lists all 29.
- Fix the microphone. Bluetooth microphones are phone quality, so pick the Mac's own microphone in the menu bar panel's Input picker. The app also warns when the room or the fans are too loud for it to hear you.
- Try a bigger model. Whisper Large v3, or the Q8 build of Turbo, means a bigger download and slower runs, and the star ratings on the AI Models page show the trade.
What it cannot do
- Intel Macs. YapToText needs Apple silicon and macOS 14 or later, and nothing is claimed for Intel Macs.
- Subtitles and speakers. File transcription gives plain text: no timestamps, no subtitles and no speaker labels.
- Translation. It types what you said in the language you said it, and OpenAI notes the turbo model is not trained for translation tasks.
- Every language equally. OpenAI says Whisper's performance varies widely depending on the language.
The offline dictation page covers what works with Wi-Fi off.
The short version
- Whisper. OpenAI's MIT-licensed speech model. YapToText runs Large v3 Turbo (Q5), the stock 574 MB whisper.cpp file.
- Local. It runs on your Mac through whisper.cpp with Metal, with no Terminal and no account.
- Everywhere. The words land at your cursor in any app, for free.
It is Whisper, on your Mac, typing wherever you are. If Whisper mishears something it should not, open an issue on GitHub. Every one gets read.
Download for free. No account, nothing locked, macOS 14 or later on Apple silicon. Source on GitHub.
Related
Sources
- https://github.com/openai/whisper (checked 2026-10-03)
- https://huggingface.co/openai/whisper-large-v3-turbo (checked 2026-10-03)
- https://github.com/ggml-org/whisper.cpp (checked 2026-10-03)
- https://huggingface.co/ggerganov/whisper.cpp/blob/main/ggml-large-v3-turbo-q5_0.bin (checked 2026-10-03)
- https://github.com/ryleighnewman/YapToText (checked 2026-10-03)
- https://apps.apple.com/us/app/yaptotext/id6786382289 (checked 2026-10-03)
Questions
Is there a Whisper app for Mac?
Yes. YapToText is a free Mac app that runs Whisper Large v3 Turbo on your Mac and types into any app. Others exist too, and the comparison page on this site sets several side by side.
Can you run Whisper locally on a Mac?
Yes. whisper.cpp runs Whisper on a Mac and is optimized for Apple silicon, and OpenAI's own package runs from Python. YapToText packages whisper.cpp and the model into an App Store app, so it runs locally with no setup.
Is Whisper free?
Yes. OpenAI released Whisper's code and model weights under the MIT License. YapToText, which runs it, is free too, with no paid tier.
Does Whisper work offline on a Mac?
Yes, when it runs locally. YapToText's App Store build ships the model, so it transcribes with no internet connection. The only network use is a model download you start yourself.
Which Whisper model is best for a Mac?
Large v3 Turbo (Q5) is the one YapToText ships, at 574 MB, because on my own dictations it heard almost exactly what the full model heard, with a third of the memory. Bigger models hear better and smaller ones are lighter, so the library lets you choose.
Does Whisper work on Intel Macs?
Yes, if you run whisper.cpp yourself: its README lists Intel and Arm Macs as supported. YapToText needs Apple silicon and macOS 14 or later, and nothing is claimed for Intel Macs.
Does Whisper run on an M4 Mac?
Yes. YapToText runs Whisper on any Apple silicon Mac with macOS 14 or later, and Apple silicon covers the M-series chips, M4 included. It needs no Python and no Terminal.