Guides

Dictation with an accent on a Mac, or in English as a second language

By Ryleigh Newman · Published October 5, 2026

Dictation with an accent works best on a Mac when you set the language you are actually speaking, or Auto-detect, give the microphone a clean signal, say each sentence in one go, and teach the app the names it keeps missing. No setting in YapToText tunes the speech model to one accent, and Apple's Dictation settings stop at language and region, so every step here is about giving the model more to work with.

I built YapToText, a free dictation app that runs OpenAI's Whisper on your Mac, because my hands make typing difficult, and I dictate every day. OpenAI says Whisper's large and diverse training data improves its robustness to accents, and its own model card also says the models perform unevenly across accents and dialects. Both are true, so this page covers what helps, what Apple Dictation offers, and what not to expect. If your words come out wrong for reasons that have nothing to do with your accent, the guide to improving dictation accuracy puts every fix in order.

Download for free. No account, nothing locked, macOS 14 or later on Apple silicon. Source on GitHub.

Steps

1. Open YapToText's AI Models page and choose the language you speak in the Language picker, or Auto-detect language if you switch.
2. Choose your microphone under Input source in Settings, General, Microphone, keep it close, and check that normal speech reaches green to yellow on the meter.
3. Say each sentence in one go, at a natural pace.
4. In History, click Show pipeline details and read the raw transcript to see what Whisper heard.
5. Add each word it keeps missing on the Dictionaries page, with the raw spelling in Heard and yours in Replace with.
6. If it still mishears you, download a bigger Whisper model from the Dictation Model Library and compare raw transcripts.

Why an accent trips up dictation

A speech model turns sound into the words it finds most likely. A voice that sounds like most of the voices it learned from gives it an easy match; a voice that sounds less like them gives it more room to guess. Your accent is not the thing to fix; the setup around it is.

Here is what OpenAI says about Whisper, the model YapToText runs:

• The training. OpenAI describes Whisper as trained on 680,000 hours of multilingual and multitask supervised data collected from the web, and says that a dataset that large and diverse leads to improved robustness to accents, background noise and technical language.
• The limit. Its model card says the models also show disparate performance on different accents and dialects of particular languages, which may include a higher word error rate across speakers of different genders, races, ages or other demographic criteria.
• Languages. OpenAI's README says Whisper's performance varies widely depending on the language.
• Names. OpenAI's Whisper prompting guide says Whisper may get uncommon proper nouns wrong, such as names of products, companies or people.

So expect most everyday English to come through, and expect the slips to cluster where the model has the least to go on: names, places, words from your own field, and words said on their own with nothing around them. Those are exactly the places the steps below work on. Whisper is one general-purpose model for everyone, and I have not tuned it for any accent; the Whisper for Mac page explains what the app runs and why.

Set the language you are speaking

The single most common cause of garbled dictation is a language setting that does not match the speaker. Tell the app which language you are speaking before you blame your accent.

Step 1: Pick the language

• You: Open YapToText's AI Models page. In Your defaults, under the dictation model, open the Language picker and choose the language you are dictating in, or Auto-detect language.
• YapToText: Hands that language to Whisper as a hint for every dictation from then on. Out of the box it uses your Mac's system language, not Auto-detect.
• Check: Dictate a sentence, then open it in History and click Show pipeline details. The Language line shows the setting that dictation used.

Step 2: Choose between a fixed language and Auto-detect

• You dictate in one language. Set it by hand. If English is your second language and your Mac's system language is your first, this matters: the default follows the system language, so pick English in the picker if English is what you dictate.
• You switch between languages. Auto-detect language lets Whisper decide the language of each dictation. If it guesses wrong for you, set the language by hand instead.
• You use two languages in different places. A mode can carry its own language in the Recording & language card of its editor, so email in one language and notes in another each get the right setting.

The picker lists 29 languages, and a few come in regional entries: English (US, UK, Australia), Spanish (Spain, Mexico), Portuguese (Brazil) and Chinese (China). Pick the entry closest to what you speak. I make no claim that a regional entry tunes Whisper to that region's accent; the language is the part that counts. Two more rules: the language is fixed when a dictation starts, so it does not switch in the middle of a sentence, and the three library models with (English) in the name hear English only. Dictating in other languages on a Mac has the full list and the per-mode setup.

What Apple Dictation offers

Apple's Dictation is built into macOS, and its language setting works by language and region. Apple lets you pick a language and a region, and switch between several while you dictate.

• You: Choose Apple menu, System Settings, then Keyboard. Go to Dictation, click Edit next to Languages, and select the language and region you speak.
• macOS: Adds it to Dictation. With more than one set up, it shows the current language next to the cursor while you dictate.
• Check: Start dictating, then click the language next to the cursor, or press the Globe key if your keyboard has one, to switch.

Apple keeps its list of Dictation languages on its macOS Feature Availability page, and says Dictation is not available in all languages or regions and that its features vary. Words it was unsure of are underlined in blue; click one to pick an alternative, or type or dictate the right word. Its own checklist for poor results asks you to speak clearly, neither too quietly nor too loud, to avoid background noise, and suggests a headset microphone in a noisy room or one with echo.

For accents, the most interesting Apple feature is not in Dictation at all. Voice Control, Apple's accessibility feature, has a custom vocabulary: Apple says you can add up to 1,000 terms in System Settings, Accessibility, Voice Control, and record how you say each one, 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.

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

Teach it the names it keeps missing

An accent rarely garbles every word. More often the same few words come out wrong the same way every time, and that kind of mistake can be taught. A word it gets wrong the same way every time can be taught; a random mistake cannot.

Step 1: Find what Whisper heard

• You: In History, click Show pipeline details on a dictation where the word came out wrong.
• YapToText: Shows the raw transcript, what Whisper heard before any fix ran.
• Check: Note the exact wrong spelling. That, not the word you meant, goes in Heard.

Step 2: Add the entry

• You: On the Dictionaries page, type the wrong spelling in Heard and your spelling 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.

Dictionary entry: For example, a colleague's surname the microphone keeps splitting in two.

Heard: okay for
Replace with: Okafor

What happens: Every okay for in a dictation is written Okafor, and Okafor is handed to Whisper as a word to listen for. If Whisper hears the name differently on different days, add each spelling, or press Generate variations on the row.

Three shortcuts save typing. Select the right spelling anywhere, tap the Quick Edit key, Right Option by default, and say add this to my dictionary. Or, right after a dictation lands, say replace, then the wrong word, then with, then the right word; fix the same mishearing twice and the Home page asks "Make this correction stick?" And with Suggest fixes from my dictations on, the default, the app lists fixes it keeps having to make, for you to add or dismiss. The Dictionaries guide covers Whole word, Case and order.

Speak in full phrases, close to the microphone

Two habits help any voice, and they help an accented voice most, because both give the model more context and a cleaner sound to match. Say the whole sentence, close to the microphone, at your normal pace.

Speak in full phrases

Have the sentence in your head, then say all of it in one breath. OpenAI's guide to its transcription API warns against splitting audio in the middle of a sentence, because that removes context and can reduce accuracy. A word said on its own gives the model nothing around it to check against, and a word inside a sentence does. There is no need to slow down or exaggerate each sound; an even, natural pace keeps the sentence together.

Say: The whole thought, once it is ready.

Please send the signed contract to the Lisbon office before Friday, and copy Ana on the email

What happens: Whisper hears Lisbon, Friday and Ana with the rest of the sentence around them, and punctuates ordinary speech like this by itself. If Ana keeps coming out as Anna, that is a job for a dictionary entry.

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

Place the microphone well

• You: Sit the way you normally use your Mac, keep the microphone close and clear of clothing, and in YapToText choose it by name under Input source in Settings, General, Microphone. Say a sentence at your normal volume.
• YapToText: Shows the Live input level meter and a rating from Excellent to Needs attention that says what to change. If the Mac's own input volume is low, it says so 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 help says Bluetooth microphones are phone quality, so keep AirPods for listening. Reduce background noise and Auto-amplify quiet speech, both on by default in the Microphone card on the Dictation page, lift a quiet voice and subtract a steady hum before Whisper hears anything, and the same card warns when the room or the fans are too loud. For Apple's Dictation, the input volume is in System Settings, Sound, Input. The microphone guide compares built-in, headset and USB microphones.

Writing in English as a second language

Many people who dictate in a second language want two things at once: their words heard correctly, and their sentences tidy. YapToText keeps those as separate steps, so you can see which one went wrong. Whisper writes what it heard; the cleanup only tidies, and only when the mode asks.

• Clean Up. Removes filler words and false starts, fixes grammar, capitalization and punctuation, and repairs words the microphone clearly misheard. It keeps your wording, tone and register, and it is told to keep the language you spoke and not translate it.
• Auto mode. On by default. For a short, plain paragraph it picks a careful pass that fixes what was misheard and rewrites nothing.
• Quick Edit. Select a sentence in any app, tap Right Option, say fix the grammar, and tap again. The result replaces your selection. The Quick Edit guide lists what else you can ask for.
• Raw Transcription. No AI at all: Whisper's words with basic punctuation. With Auto mode on, press 2 while you talk to use it for one dictation.
• Translation, as a separate step. To turn finished text into another language, select it, tap the Quick Edit key and say the language, or run the Translate to English AI Action. Dictation itself never translates.

One thing to know before you start: the app's own screens are in English only, and the default spoken commands are English phrases such as question mark or insert comma. You can edit any command, or add trigger phrases of your own, on the Commands page. If you want to read every dictation before it lands, turn on Review before inserting in Settings, Advanced, Text output. The modes guide explains what each mode may change.

If it keeps mishearing you

Climb this ladder, smallest step first:

1. Check the language. The Language picker on the AI Models page sets the default, and a mode with its own language overrides it. The Language line in a dictation's pipeline details shows what was used.
2. Check the model. A model with (English) in its name hears English only. Switch back to Whisper Large v3 Turbo (Q5), the multilingual default, for anything else.
3. Read the raw transcript. If the word is wrong there, Whisper misheard it. If it is right there and wrong in the final text, the cleanup changed it, and Raw Transcription gives you Whisper's words.
4. Teach the repeat offenders. Add each word that is wrong the same way every time to a dictionary.
5. Fix the signal. Move the microphone closer, pick it by name, quiet the room, and watch the meter.
6. Try a bigger model. In the Dictation Model Library, download 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. Click the circle at the start of its row to use it from the next dictation, then compare raw transcripts in History.
7. Tell me. Open an issue on GitHub with the language, a sample sentence and the raw transcript.

The app sums up the model trade in its own words: bigger models hear better, and smaller ones are lighter on battery and memory. A bigger model may change little for your voice, or a lot, and only your own History can say. The AI models guide lists every model and size. Fix the language and the sound before you reach for a bigger model.

What not to expect

• Perfect results for any accent. I will not claim that any app on this page handles any accent perfectly, mine included. Whisper is a general-purpose model, and OpenAI itself says it performs unevenly across accents and dialects.
• A number. I have not run an accuracy test for any accent, on Apple's Dictation or on YapToText, so this page has no percentages. A few minutes with your own sentences is the test that counts.
• Training to your voice. Nothing in YapToText retrains Whisper for one person. Dictionaries teach spellings and prime the model with your words, which is different from learning your accent. Apple's Dictation has no word list, and Voice Control's vocabulary works only while Voice Control replaces standard Dictation.
• A switch mid-sentence. YapToText sets the language when a dictation starts. Apple's Dictation is the one that switches as you talk.
• Help with speech that is hard to understand. The audio work in YapToText is for quiet voices and noisy rooms. Dictation by condition is honest about where that falls short.
• An Intel Mac. YapToText needs an Apple silicon Mac with macOS 14 or later.

Settings can give your voice a fair hearing; they cannot promise a perfect one.

The short version

• Language first. Pick the language you are speaking, or Auto-detect if you switch, in YapToText's Language picker or Apple's Languages list.
• Then the sound. Microphone close and picked by name, room quiet, whole sentences at a natural pace.
• Then the words. Names it keeps missing go in a dictionary, or in Voice Control's vocabulary on Apple's side.
• Then the model. A bigger Whisper model, tested against your own History.

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

Set it up for the voice you have, then let your own dictations tell you how well it hears you. If your accent or language trips it up in a way this page does not cover, open an issue on GitHub with a sample sentence. Every one gets read.

Download 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.

Related

• How to improve Mac dictation accuracy: ten fixes, biggest first
• Mac dictation in other languages, and in more than one
• Dictionaries: teach YapToText your names and jargon
• The best microphone for dictation on a Mac, and how to turn it on
• The Whisper and cleanup models YapToText runs on your Mac
• Whisper hallucinations: why Whisper types words nobody said, and how to stop it
• Apple Dictation vs Whisper on a Mac: the speech engines, not the apps
• Spanish dictation on a Mac: set it up in Apple Dictation or YapToText
• Dictating in two languages on a Mac: switch as you talk, or set it per dictation
• Teach Mac dictation new words: names, jargon, brands

Sources

• https://openai.com/index/whisper/ (checked 2026-10-05)
• https://github.com/openai/whisper/blob/main/model-card.md (checked 2026-10-05)
• https://github.com/openai/whisper (checked 2026-10-03)
• https://developers.openai.com/cookbook/examples/whisper_prompting_guide (checked 2026-10-05)
• https://developers.openai.com/api/docs/guides/speech-to-text (checked 2026-10-05)
• https://support.apple.com/guide/mac-help/use-dictation-mh40584/mac (checked 2026-10-03)
• https://www.apple.com/macos/feature-availability/ (checked 2026-10-03)
• https://support.apple.com/guide/mac-help/if-dictation-on-mac-doesnt-work-as-expected-mchlc480652b/mac (checked 2026-10-03)
• https://support.apple.com/guide/mac-help/change-the-sound-input-settings-mchlp2567/mac (checked 2026-10-03)
• https://support.apple.com/guide/mac-help/use-a-custom-vocabulary-mchl3eb7b79a/mac (checked 2026-10-05)
• https://support.apple.com/guide/mac-help/use-voice-control-commands-mh40719/mac (checked 2026-10-03)
• https://apps.apple.com/us/app/yaptotext/id6786382289 (checked 2026-10-03)

Questions