Guides

How to improve Mac dictation accuracy: ten fixes, biggest first

By · Published

To make dictation more accurate on a Mac, fix the sound first: bring the microphone closer, quiet the room and set the input volume so your voice arrives clearly. Then match the language to the one you speak, talk in whole phrases, and teach it the names it keeps getting wrong. Those six fixes are the ones to try first, in Apple's Dictation and in YapToText alike.

Below are ten fixes in order, biggest first, each tied to the setting that does it. The last four are YapToText's: a bigger speech model, the raw transcript that shows what was actually heard, and two voice helpers that are on by default. I built YapToText, a free dictation app, because my hands make typing difficult, and I dictate every day, quietly, often next to a running 3D printer. If dictation does nothing at all, Mac dictation not working is the page you want. This one is for when it works but keeps getting words wrong.

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: 10 sections
  1. Steps
  2. The ten fixes, biggest first
  3. Fixes 1 to 3: get a clean signal
    1. 1. Bring the microphone closer, and pick it by name
    2. 2. Quiet the room
    3. 3. Set the input volume
  4. Fixes 4 and 5: the right language, in whole phrases
    1. 4. Tell it the language you speak
    2. 5. Speak in full phrases
  5. Fix 6: teach it names and jargon
    1. Can you train Mac dictation?
    2. Teach YapToText with Dictionaries
  6. Fix 7: use a bigger speech model
    1. Step 1: Download a bigger model
    2. Step 2: Select it and compare
  7. Fixes 8 to 10: see what it heard, and leave the helpers on
    1. 8. Read the raw transcript
    2. 9. Leave Auto-amplify quiet speech on
    3. 10. Leave Reduce background noise on
  8. What no setting can fix
  9. The short version
  10. Questions

Steps

  1. Bring the microphone closer and choose it by name under Microphone source, or under Input source in YapToText.
  2. Quiet the room, or use a headset microphone if it is noisy or echoes.
  3. Raise the input volume in System Settings, Sound, Input, or raise Input boost in YapToText until normal speech reaches green to yellow.
  4. Under Dictation in Keyboard settings, click Edit next to Languages and select the language and region you speak.
  5. Speak in whole phrases, and add names that keep coming out wrong to YapToText's Dictionaries.
  6. In YapToText, open History, click Show pipeline details and read the raw transcript to see what was actually heard.

The ten fixes, biggest first

Most wrong words come from one of three places: what the microphone caught, what the model expected to hear, or how the words were said. Work down the list in order, and stop when the words come out right. Fix what the microphone hears before you fix what the model knows.

Apple's side from its support pages as I read them on October 3, 2026, and Voice Control's vocabulary page on October 5, 2026. YapToText's side from the app's own settings.

#FixApple DictationYapToText
1Microphone distance and choiceMicrophone source, under Dictation in Keyboard settingsInput source, in Settings, General, Microphone
2Room noiseAvoid background noise; a headset for noise or echoA warning on the Dictation page when the room is too loud
3Input volumeSystem Settings, Sound, InputInput boost, and a warning when the Mac's input volume is low
4The right languageLanguages, under DictationThe Language picker on the AI Models page
5Full phrasesA speaking habitA speaking habit
6Names and jargonNo word list under Dictation; Voice Control has a custom vocabularyThe Dictionaries page
7A bigger speech modelNot a choice; a newer model comes to some Macs with macOS 27The Dictation Model Library on the AI Models page
8See what was actually heardBlue underlines on words it was unsure ofThe raw transcript in History, or Raw Transcription
9Auto-amplify quiet speechNo setting under DictationOn by default, Dictation page, Microphone card
10Reduce background noiseNo setting under DictationOn by default, Dictation page, Microphone card

The Dictation settings Apple lists in Keyboard settings are the switch, Languages, Microphone source, Shortcut and Auto-punctuation, which is why the Apple column runs thin toward the bottom.

Fixes 1 to 3: get a clean signal

A speech model can only write down what reaches it. A voice that arrives faint, far away or buried in a fan gives any model, Apple's or Whisper, less to work with. A clean signal fixes more words than any setting after it.

1. Bring the microphone closer, and pick it by name

Apple's own checklist asks you to sit the way you normally use your Mac, to keep clothing and your body clear of the microphone, and to speak clearly, neither too quietly nor too loud. Then make sure dictation is listening to the microphone you are actually talking into. In Keyboard settings, under Dictation, Microphone source set to Automatic picks the device your Mac thinks you are most likely to use, which may be a headset across the room. In YapToText, choose it under Input source in Settings, General, Microphone. YapToText's help says plainly that Bluetooth microphones are phone quality, so keep AirPods for listening. The guide to choosing a microphone for dictation compares built-in, headset and USB microphones and shows how to test two of them side by side.

2. Quiet the room

Apple's advice is to avoid background noise, and in a noisy room or one with a lot of echo, it suggests a headset microphone may help. Turn off what you can: a fan, a radio, a video in another tab. YapToText's Pause music and video while dictating, on by default in the Recording card, pauses playback when a dictation starts and resumes it when the dictation ends. Its Microphone card on the Dictation page also warns when the room or the fans are too loud for the microphone to hear you, so you know to move instead of guessing.

3. Set the input volume

For Apple's Dictation, choose Apple menu, System Settings, click Sound, then Input, select your microphone and drag the input volume slider. Apple's own example of when to raise it is someone who speaks softly. YapToText has its own check:

  • You: Open Settings, General, Microphone, and say a sentence at your normal volume.
  • YapToText: Shows the Live input level meter, and a rating underneath from Excellent to Needs attention that says what to change. If the Mac's own input volume is low, the app says so, with the percentage, and offers a Raise to 80% button.
  • Check: Normal speech reaches the green-to-yellow zone without sitting in the red. If it does not, raise Input boost on the same card.
YapToText's Settings, General tab: the Shortcuts card, the Auto mode card, and the Microphone card with a live input level meter.
Figure 1. Settings, General, Microphone in YapToText: Input source names the microphone it hears, and the Live input level meter shows whether your voice reaches it at a useful level.

If the meter stays flat whatever you do, the problem is not accuracy any more, and the troubleshooting steps for silence come first.

Fixes 4 and 5: the right language, in whole phrases

4. Tell it the language you speak

When words come out wrong, check the language before anything else in the settings. Check the language setting before you blame your accent.

  • Apple Dictation. Under Dictation in Keyboard settings, click Edit next to Languages and select the language and the region you speak. With more than one set up, switch while dictating by clicking the language shown next to the cursor, or by pressing the Globe key if your keyboard has one.
  • YapToText. The Language picker sits under the dictation model on the AI Models page. It starts on your Mac's system language and offers Auto-detect plus 29 languages, so if you dictate in a language other than your system one, pick it there. The three English Whisper models in the library hear English only, so they are the wrong choice for anything else.

A mode can carry its own language too, in the Recording & language card of its editor, which suits someone who writes email in one language and notes in another. Dictating in other languages on a Mac covers both apps in detail.

5. Speak in full phrases

Say the whole thought, at an even pace, instead of feeding it one word at a time. OpenAI's own guide to its transcription API warns against splitting audio in the middle of a sentence, because that removes context and can reduce accuracy. That advice is about cutting files, but the same idea applies to how you talk: a word said on its own gives the model nothing around it to check against.

Say: The whole sentence in one breath, after you have it in your head.

Can you move the review with Dr. Okafor to Thursday at 3:30, question mark

What happens: In YapToText, the sentence lands with the question mark, because question mark is a two-word name said at the end of a clause. Said word by word, with long gaps, the same sentence gives the model far less to go on.

Whole phrases help YapToText after the transcript too. Under Auto mode, six words or fewer stay exactly as spoken, so a long thought chopped into scraps also misses the careful pass that fixes misheard words. Think first, then say the whole sentence.

Fix 6: teach it names and jargon

Names, product words and the jargon of your job are where every speech model slips. OpenAI's Whisper prompting guide says Whisper may get uncommon proper nouns wrong, such as names of products, companies or people. Mine is the classic case: the microphone hears Riley, and I spell it Ryleigh. A word it gets wrong the same way every time can be taught; a random mistake cannot.

Can you train Mac dictation?

Not Apple's Dictation itself: none of its settings is a word list. Voice Control, Apple's separate accessibility feature, has one. Apple says you can add up to 1,000 vocabulary 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 a custom vocabulary. The catch is that while Voice Control is on, standard Dictation is not available, and you dictate through Voice Control instead. The Voice Control versus Dictation guide weighs that trade.

Teach YapToText with Dictionaries

  • You: In History, click Show pipeline details on a dictation where the word came out wrong and note its spelling in the raw transcript. On the Dictionaries page, type that spelling in Heard and yours in Replace with, then press Add.
  • YapToText: Writes your spelling from the next dictation on, and also hands it to Whisper as a word to listen for, so the fix works before the transcript as well as after it.
  • Check: Dictate the word again, and it lands spelled your way.

Two shortcuts skip the typing. Select the right spelling anywhere, tap Right Option, say add this to my dictionary, and tap again. Or fix the same mishearing twice with replace, then the wrong word, then with, then the right word, and the Home page asks “Make this correction stick?” The Dictionaries guide covers Whole word, Case, order and the limits.

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 pairs what the speech model hears with the spelling you want, and Suggested for you offers the fixes that keep coming back in your own dictations.

Fix 7: use a bigger speech model

YapToText ships with Whisper Large v3 Turbo (Q5), 574 MB, as its speech model. The AI Models page sums up the trade in the app's own words: bigger models hear better, and smaller ones are lighter on battery and memory. Try a bigger model only after the sound and the words are right.

Step 1: Download a bigger model

  • You: On the AI Models page, in the Dictation Model Library, press the download button on a larger Whisper model: Large v3 Turbo (Q8) at 834 MB, Large v3 Turbo at 1.5 GB, Large v3 (Q5) at 1.1 GB, or Large v3 at 3.0 GB, the largest and slowest.
  • YapToText: Fetches it from Hugging Face with a progress bar, the one kind of network use the app has.
  • Check: The row says Installed.

Step 2: Select it and compare

  • You: Click the circle at the start of its row, then dictate a few sentences you often say.
  • YapToText: Uses that model from the next dictation.
  • Check: In History, compare the raw transcripts against your older ones. Keep the bigger model only if it fixes words you care about.
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 AI Models page lists each speech model with its size and star ratings for accuracy and speed, so you see the cost before you download.

Be honest with yourself about the result. I picked the 5-bit build as the default because, on my own dictations, it heard almost exactly what the full model heard with a third of the memory, so a bigger model may change little. A mode can carry its own speech model, and the Energy page can switch to a lighter one on battery. The models guide lists every model and size.

Apple's Dictation model is not a choice you make. Apple lists improved accuracy in Dictation, from a new on-device model, as part of the next generation of Apple Intelligence 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. Apple Intelligence and dictation has the details.

Fixes 8 to 10: see what it heard, and leave the helpers on

8. Read the raw transcript

Before you change anything else, find out where the wrong word came from. In YapToText, every dictation in History has Show pipeline details, which shows the raw transcript, what Whisper heard before any cleanup, above the final text.

A YapToText History entry with pipeline details open: raw transcript, cleaned transcript, mode, delivery, target app and timings.
Figure 4. Show pipeline details puts the raw transcript above the final text, so you can tell a word the model misheard from a word the cleanup changed.
  • Wrong in the raw transcript. The model misheard it. Fixes 1 to 7 are the ones that help.
  • Right in the raw transcript, wrong in the final text. The cleanup, a dictionary entry or a command changed it. While you talk, press 2 for Raw Transcription, which with Auto mode on uses no AI model for that one dictation, or choose Copy Raw Transcript on the History entry. The modes guide explains what each mode may change.

Apple's Dictation has its own version of this: words it was unsure of are underlined in blue, and you can click one and choose an alternative, or type or dictate the right word. Know whether the model or the cleanup made the mistake before you fix it.

9. Leave Auto-amplify quiet speech on

On by default, in the Microphone card on YapToText's Dictation page. It measures your voice against the room instead of against a fixed level, so a voice that cannot project still gets lifted. The same card warns when your signal drops well below your own normal and names the likely cause.

10. Leave Reduce background noise on

Also on by default, on the same card. When a machine hums near the microphone, the app subtracts that steady sound before Whisper transcribes, and clean audio skips the step. Both helpers run on your Mac. They are last on this list because they are already working; the fix is only to check that neither was switched off while you tested something.

YapToText's Dictation settings page on a Mac: the Keys, Recording, Microphone and Delivery cards.
Figure 5. Reduce background noise and Auto-amplify quiet speech share the Microphone card on the Dictation page, and both are on out of the box.

What no setting can fix

  • Sound that was never caught. No boost, model or noise reduction can recover words the microphone did not pick up.
  • Speech that is hard to understand. Whisper is a general-purpose model, and nothing in YapToText changes how it handles slurred speech. Dictation by condition says honestly where it may fall short.
  • A promise of a number. I have not run an accuracy test on Apple's Dictation or on YapToText, so this page has no percentages. Your own History is the test that counts.
  • Apple's model. You cannot swap the model inside Apple's Dictation or give it a word list.
  • Intel Macs. YapToText needs an Apple silicon Mac with macOS 14 or later, and nothing is claimed for Intel Macs.

Settings can lift a clear voice; they cannot invent a word that was never heard.

The short version

  • Sound first. Microphone close and picked by name, the room quiet, the input level in the green.
  • Then the words. The right language, whole phrases, and your names in Dictionaries or Voice Control's vocabulary.
  • Then the model. A bigger Whisper model in YapToText, tested against your own History.
  • Look before you fix. The raw transcript says whether the model or the cleanup got it wrong.

YapToText is free on the Mac App Store, and everything on this page runs on your Mac.

Most wrong words start at the microphone, so start there too. If a fix here does not match what you see, 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

Why is my Mac dictation not accurate?

Usually the sound: the microphone is too far away, the wrong one is selected, the room is noisy or the input volume is low. After that, check that the language under Dictation in Keyboard settings matches the one you speak.

How do I make dictation more accurate on a Mac?

Bring the microphone closer and pick it by name, quiet the room, raise the input volume, match the language, and speak in whole phrases. Teach it names it keeps getting wrong, and in YapToText try a bigger Whisper model.

Can you train Mac dictation to learn new words?

Apple's Dictation has no word list. Voice Control lets you add up to 1,000 custom vocabulary terms, but standard Dictation is not available while it is on. YapToText's Dictionaries teach it your spellings and also prime the speech model.

Why does dictation keep getting words wrong on my Mac?

If the same word is wrong the same way every time, the model does not know it, so teach it. If different words go wrong, look at the microphone, the room and the input volume first.

Does a better microphone improve dictation accuracy?

It can, because the model can only write what reaches it. Apple suggests a headset microphone in a noisy room or one with echo. Test with your own voice: YapToText keeps each take in History to compare.

Does a bigger Whisper model make dictation more accurate?

YapToText's AI Models page says bigger models hear better and smaller ones are lighter on battery and memory. The default, Whisper Large v3 Turbo (Q5), is a good start; compare a bigger one against your own History before keeping it.