Guides

Apple Dictation vs Whisper on a Mac: the speech engines, not the apps

By · Published · Updated

Apple Dictation vs Whisper comes down to who runs the engine and where. Apple Dictation is the speech recognition built into macOS, processed on your Mac or sent to Siri servers depending on your Mac and your language. Whisper is OpenAI's open speech model, and when an app or whisper.cpp runs it locally, it runs entirely on your Mac. Apple's needs nothing installed and stops after 30 seconds of silence; Whisper has no silence rule of its own, takes your words only through the app that runs it, and transcribes files as well as live speech. I found no source that compares their accuracy directly, so this page does not rank them on it.

This page compares the engines, not the apps built on them. If you are choosing between Apple's built-in dictation and a full app, Mac dictation vs Wispr Flow covers that. I built YapToText, a free app that runs Whisper on the Mac, so weigh what I say about it with that in mind. Apple facts come from support.apple.com and apple.com, and Whisper facts from OpenAI's and whisper.cpp's own pages, read on October 3 and October 5, 2026.

Download YapToText on the Mac App StoreDownload for free. No account, nothing locked, macOS 14 or later on Apple silicon. Source on GitHub.

On this page: 11 sections
  1. Steps
  2. Two engines, and what runs them
  3. Side by side
  4. Where each engine runs
    1. Check where Apple Dictation runs on your Mac
  5. Languages, punctuation and the silence stop
    1. Languages
    2. Punctuation
    3. The silence stop
  6. Custom words and file transcription
    1. Teaching each engine your words
    2. Transcribing a file
  7. Accuracy: what Apple and OpenAI publish
    1. Step 1: Write down five sentences you really say
    2. Step 2: Dictate them with Apple Dictation
    3. Step 3: Dictate them with Whisper, raw
    4. Step 4: Add the pause and the name
  8. Try Whisper on your Mac
    1. With YapToText
    2. With whisper.cpp
  9. What each one cannot do
    1. Apple Dictation
    2. Whisper
  10. The short version
  11. Questions

Steps

  1. Write down five sentences you really say, with at least one name that usually comes out wrong.
  2. Dictate them with Apple Dictation and count the words you would have to fix.
  3. Dictate the same sentences in YapToText with Raw Transcription, so only Whisper is involved, and count the fixes the same way.
  4. Pause for 30 seconds mid-sentence in each, then teach the name to Voice Control or YapToText Dictionaries and try again.

Two engines, and what runs them

The two names describe different kinds of thing, which is why the comparison gets muddled. Apple Dictation is a feature you switch on; Whisper is a model something has to run.

  • Apple Dictation. Built into macOS. You turn it on in System Settings, Keyboard, Dictation, and start it with the Microphone key, your Dictation shortcut or Edit, Start Dictation. You cannot swap the model inside it. Apple lists improved accuracy in Dictation from a new on-device model in macOS 27 Golden Gate, on Mac models with M3 or later and at least 12 GB of unified memory, marked as available in English.
  • Whisper. OpenAI calls it a general-purpose speech recognition model trained on a large dataset of diverse audio, and it released the code and weights under the MIT License. Turbo, the version most Mac apps use, is what OpenAI calls an optimized version of large-v3 with faster transcription and a minimal degradation in accuracy.
  • What runs Whisper. On a Mac, usually whisper.cpp, an MIT-licensed C and C++ engine whose README calls Apple silicon a first-class citizen and runs inference on the GPU through Metal. YapToText builds a copy of whisper.cpp into the app and ships the stock Whisper Large v3 Turbo (Q5) file.

Whisper vs Siri dictation is the same question, because Apple's own wording in Keyboard settings is about Siri servers. Voice Control is a separate Apple feature; while it is on, standard Dictation is not available.

Side by side

Checked October 3, 2026, on support.apple.com, OpenAI's Whisper README and whisper.cpp's README, and October 5, 2026, for Voice Control's vocabulary page and OpenAI's Whisper prompting guide.

Apple DictationWhisper
What it isApple's built-in DictationOpenAI's speech model, run by an app or whisper.cpp
Where it runsOn the Mac or on Siri servers; the text under Dictation in Keyboard settings says whichWherever it is run; locally, on the Mac
Without internetOn Apple silicon in supported languages, per Apple's Sonoma guideYes, once the model file is on the Mac
LanguagesNot every language or region; switch languages while you dictateMultilingual models, with performance that varies widely by language; YapToText lists 29 plus auto-detect
SilenceStops after 30 seconds without speechNo rule of its own; the app decides
PunctuationCommas, periods and question marks on their own in supported languages, plus spoken namesPunctuates ordinary speech itself in YapToText; spoken names for the rest
Your own wordsNo word list; Voice Control holds up to 1,000 termsA prompt can suggest spellings; YapToText's Dictionaries prime it
Audio filesNot a Dictation job; Voice Memos and Notes transcribe their own recordingsYes, with whisper-cli or an app
Choice of modelNonetiny to large-v3, turbo and quantized files
MacsApple silicon and Intel; Intel sends your words to Applewhisper.cpp: Intel and Arm; YapToText: Apple silicon, macOS 14 or later
CostBuilt into macOSFree, MIT; YapToText is free too

Apple decides where Dictation runs; with Whisper, whoever runs it decides, and on a Mac that can be you.

Where each engine runs

Apple's Mac User Guide for macOS Sonoma says that on a Mac with Apple silicon, dictation requests are processed on your device for supported languages, with no internet connection required, and that on an Intel-based Mac, or in a language without on-device support, your dictated words are sent to Apple. The current guide turns that into a check, and its troubleshooting page adds that you may need an internet connection. Apple also stores audio of your Siri and Dictation interactions if you choose Share Audio Recordings, which you can change later in Privacy & Security, Analytics & Improvements.

Check where Apple Dictation runs on your Mac

  • You: Choose Apple menu, System Settings, click Keyboard, and read the text below Dictation.
  • macOS: Says whether your voice inputs are processed on your Mac and not sent to Siri servers.
  • Check: To be sure, turn Wi-Fi off and dictate a sentence. The offline dictation page has the full test.

Whisper has three pieces, the model file, the engine and your audio, and all three can sit on your Mac. The internet is needed only to download the model file once, and YapToText's App Store build has the file inside, so it skips even that. Local Whisper changes where your voice goes, not how well it is heard. The local Whisper guide lists what each model size takes in disk space and memory.

Languages, punctuation and the silence stop

Languages

Apple says Dictation is not available in all languages or regions and that its features vary, and it keeps the list on its macOS Feature Availability page. Its strength is switching: with several languages set up under Dictation in Keyboard settings, you click the language next to the cursor, or press the Globe key, while you dictate. OpenAI says Whisper's performance varies widely depending on the language, and its tiny, base, small and medium models also come as English-only .en files. YapToText sets one language per dictation, from 29 plus Auto-detect, so Apple is the one to use if you switch languages mid-sentence. The languages guide lists all 29.

Punctuation

In supported languages, Apple Dictation adds commas, periods and question marks on its own, which it calls Auto-punctuation, and you can say a mark's name, such as comma or new paragraph. In YapToText, Whisper punctuates ordinary speech by itself, and spoken names cover the marks it cannot guess.

Say: In YapToText, at the end of a clause.

is it working now, question mark

What happens: You get "is it working now?" One-word names such as comma need insert in front while that setting is on, which is the default. The spoken punctuation guide lists every name.

The silence stop

Apple says you can dictate text of any length without a timeout, but Dictation stops automatically when no speech is detected for 30 seconds, and its guide describes no setting that changes it. Whisper has no stop of its own; the app around it decides. In YapToText, Auto-stop after silence and Maximum length are both off by default, so a pause never ends a dictation, and a safety stop ends one dictation at 30 minutes and still saves it. The 30-second stop belongs to Apple Dictation, not to speech recognition. The 30-second limit page explains the workarounds.

YapToText's Dictation settings page on a Mac: the Keys, Recording, Microphone and Delivery cards.
Figure 1. The Recording card in YapToText: Auto-stop after silence and Maximum length are off by default, so the silence rule Apple Dictation has does not apply.

Custom words and file transcription

Teaching each engine your words

Apple Dictation has no word list. Words it was unsure of are underlined in blue, and you click one to choose an alternative. Voice Control, Apple's separate accessibility feature, does have a custom vocabulary: Apple says you can add up to 1,000 terms in System Settings, Accessibility, Voice Control, record how you say each one, and add a selected word by saying Add to vocabulary, in languages that support it. While Voice Control is on, standard Dictation is not available, so you dictate through Voice Control instead.

Whisper has no list either. OpenAI's Whisper prompting guide says it may get uncommon proper nouns wrong, such as names of products, companies or people, and that a prompt with the right spellings can help, though it calls these techniques not especially reliable. YapToText does it for you: each entry on the Dictionaries page fixes the spelling and also hands the word to Whisper as a word to listen for.

YapToText's Dictionaries page: suggested fixes such as thursday to Thursday, and the General dictionary with iPhone and Wi-Fi.
Figure 2. The Dictionaries page in YapToText: each entry fixes how a word is written and also primes Whisper to listen for it.

The guide to improving dictation accuracy works through both engines, fix by fix.

Transcribing a file

Apple Dictation listens to a microphone while your cursor waits; it does not take a file. Apple's file-shaped options are Voice Memos, which shows a transcript of its own recordings on macOS 15 or later with Apple silicon, where available, and Notes, which transcribes audio you record in a note on a Mac with M1 or later. Whisper takes files by design: whisper.cpp's whisper-cli transcribes 16-bit WAV files, and YapToText's Transcribe a file card takes any audio or video file macOS can decode, as plain text. For a recording you already have, the engine you want is Whisper. The free audio transcription guide has the steps.

Accuracy: what Apple and OpenAI publish

There is no published head-to-head test from Apple or OpenAI that I can point you to, and I have not run an accuracy test myself. So there is no number here. What the sources do say:

  • Apple. It lists improved accuracy from a new on-device Dictation model in macOS 27, on M3 or later Macs with at least 12 GB of memory, in English. Its own example of a word Dictation can get wrong is flour for flower.
  • OpenAI. Turbo trades a minimal degradation in accuracy for speed against large-v3, and performance varies widely by language.
  • A third party. Aqua publishes its own benchmark of technical vocabulary and says its model beats Whisper Large v3 on it. That compares Whisper with Aqua's model, not with Apple, and it is Aqua's test.

Your own sentences are the only fair test, and it takes ten minutes.

Step 1: Write down five sentences you really say

  • You: Pick sentences you send every week, with at least one name that usually comes out wrong.
  • Check: Read them aloud once so the pace is natural.

Step 2: Dictate them with Apple Dictation

  • You: Click into a note, press your Dictation shortcut, say the five sentences and press Esc.
  • macOS: Types them, with words it was unsure of underlined in blue.
  • Check: Count the words you would have to fix.

Step 3: Dictate them with Whisper, raw

  • You: In YapToText, tap Right Command, press 2 for Raw Transcription while you talk, say the same sentences and tap again.
  • YapToText: Types exactly what Whisper heard, with no cleanup model involved.
  • Check: Count the fixes the same way. Comparing raw against raw keeps the cleanup out of the score.

Step 4: Add the pause and the name

  • You: Pause for 30 seconds in the middle of a sentence in each, then teach the name to Voice Control's vocabulary or YapToText's Dictionaries and try again.
  • Check: Apple Dictation ends at the pause; YapToText keeps listening until you tap.

Try Whisper on your Mac

There are two ways, and they end in different places. An app puts Whisper's words at your cursor; whisper.cpp prints them in Terminal.

With YapToText

  • You: Get YapToText from the Mac App Store, allow the microphone, and switch it on in System Settings, Privacy & Security, Accessibility.
  • YapToText: Arrives with Whisper Large v3 Turbo (Q5) inside, so there is nothing else to download.
  • Check: Click into any text field, tap Right Command, talk, and tap again. The words land at your cursor.
The AI Models page in YapToText: Whisper Large v3 Turbo (Q5) and Phi-3.5 Mini Instruct marked Included, model library below.
Figure 3. The Whisper model YapToText runs, under Your defaults on the AI Models page, with ten more Whisper models in the library below.

With whisper.cpp

Terminal: The commands from whisper.cpp's README, with the 5-bit turbo model. They assume Git and CMake.

git clone https://github.com/ggml-org/whisper.cpp.git
cd whisper.cpp
sh ./models/download-ggml-model.sh large-v3-turbo-q5_0
cmake -B build
cmake --build build -j --config Release
./build/bin/whisper-cli -m models/ggml-large-v3-turbo-q5_0.bin -f samples/jfk.wav

What happens: The script downloads the model from Hugging Face, CMake builds whisper-cli, and it prints the sample recording as text in Terminal. Your own audio needs converting to 16-bit WAV first.

The local Whisper guide explains each command, and the Whisper for Mac page covers the model inside YapToText.

What each one cannot do

Apple Dictation

  • Keep listening through a pause. It stops after 30 seconds without speech.
  • Learn a word list. Only Voice Control has one, and it replaces Dictation while it is on.
  • Take a file. Voice Memos and Notes transcribe only their own recordings.
  • Let you choose the model, or where it runs. Your Mac and your language decide.

Whisper

  • Run without something around it. It is a model; you need whisper.cpp, OpenAI's Python package or an app.
  • Switch languages mid-sentence in YapToText. The language is set when a dictation starts.
  • Run YapToText on an Intel Mac. YapToText needs Apple silicon and macOS 14 or later, though whisper.cpp itself lists Intel Macs as supported.
  • Translate with turbo. OpenAI says the turbo model is not trained for translation.

Each engine is limited by what surrounds it: macOS for Apple, and the app you choose for Whisper.

The short version

  • Apple Dictation. Built in, on your Mac or Siri servers, switches languages as you talk, stops after 30 seconds of silence, no word list.
  • Whisper. OpenAI's open model, fully on your Mac when run locally, no silence stop of its own, transcribes files, and learns words through the app.
  • Accuracy. No source compares them directly, so test both on your own sentences.
  • Trying Whisper. YapToText for typing into any app, whisper.cpp for Terminal.

Pick the engine by where it runs and how it stops; let your own sentences settle accuracy.

If something here does not match what you see on your Mac, open an issue on GitHub. Every one gets read.

Download YapToText on the Mac App StoreDownload for free. No account, nothing locked, macOS 14 or later on Apple silicon. Source on GitHub.

About the author. Ryleigh Newman is a developer whose hands make typing difficult and who dictates every day with YapToText, the app this site describes. Facts about other apps come from their own sites, with the date they were read. More about the project.

Sources

Questions

Is Whisper better than Apple Dictation?

No source I found compares their accuracy directly, so test both on your own sentences. Whisper, run locally, always stays on your Mac, has no 30-second silence stop and transcribes files. Apple Dictation needs nothing installed and switches languages while you talk.

Does Apple Dictation use Whisper?

Apple's guides that I read do not say so; they describe Apple's own Dictation, and Apple lists a new on-device Dictation model in macOS 27. YapToText runs Whisper itself and does not use Apple's Dictation at all.

What is the difference between Whisper and Siri dictation?

Siri dictation usually means Apple Dictation, which runs on your Mac or on Siri servers depending on your Mac and language. Whisper is OpenAI's open speech model, and when an app such as YapToText runs it locally, your voice never leaves the Mac.

Does Whisper work offline on a Mac?

Yes, when it runs locally. After the model file is downloaded once, whisper.cpp needs no connection, and the App Store build of YapToText ships with the model inside.

Can Apple Dictation transcribe an audio file?

No. Dictation listens to your microphone. Voice Memos and Notes transcribe recordings made in those apps on supported Macs, and for a file you already have, you need something like whisper.cpp or YapToText.

Can I add custom words to Apple Dictation?

Not to Dictation itself. Voice Control has a custom vocabulary of up to 1,000 terms, but standard Dictation is not available while Voice Control is on. YapToText's Dictionaries teach Whisper your words.