TL;DR
- Mac Dictation can work well for non-native English speakers, but the public accuracy data is thin.
- The best 2025 study on non-native speech tested Whisper and AssemblyAI, not Apple's built-in Dictation.
- Pauses, fillers, self-corrections, mic distance and language settings change results more than most people expect.
- Run a 120-word test on your own voice before you trust any dictation app.
1.14 billion
English speakers learned it as a second language, about three-quarters of the 1.53 billion total (Ethnologue 2025)
A Detailed Look at How Many People Speak English Globally (Rosetta Stone)0.054
Best mean Match Error Rate on non-native read speech (Whisper), McGuire 2025. Lower is better
Automatic Speech Recognition for Non-Native English: Accuracy and Disfluency Handling30 seconds
Silence before Mac Dictation stops by itself. Apple says Dictation has no set length limit
Dictate messages and documents on Mac (Apple Support)12GB
Minimum unified memory for the macOS 27 on-device dictation enhancement, on M3 or later Macs
Dictate messages and documents on Mac (Apple Support)Dictation on a Mac works for many non-native English speakers, but the accuracy data is thinner than most product pages suggest. The best 2025 study we found on non-native speech put the top systems close to human-level accuracy on read sentences, yet it did not test Apple's built-in Dictation.
That gap matters. You need to know what a number measures before you trust it. This guide covers what the studies show, why results swing from one speaker to the next, how to set up Dictation for English on your Mac, and how to run a short test on your own voice.
What the accuracy studies actually show
Most people want one number. They want the error rate for someone who speaks English as a second language, using Apple's tool. That number does not exist in public research yet. What does exist is smaller and more specific. The table lists the studies we could open.
| Study | What it tested | Result | What it does not tell you |
|---|---|---|---|
| McGuire, arXiv 2503.06924 (2025), read speech | 2,400 single-sentence recordings from 24 non-native speakers, L2-ARCTIC corpus | Best mean Match Error Rate: 0.054 (Whisper), 0.056 (AssemblyAI) | Uses MER, not WER. No native-speaker baseline. Apple Dictation not tested. |
| McGuire, arXiv 2503.06924 (2025), spontaneous speech | Narrative recordings from 22 speakers | Best mean MER: 0.063 (RevAI) | Results varied by system and by disfluency type. |
| Koenecke et al., PNAS 2020 | Five commercial systems from Amazon, IBM, Google, Microsoft and Apple | Average word error rate of 35% for Black speakers, 19% for white speakers | Concerns race and dialect, not non-native speakers. |
| Kuhn et al., arXiv 2408.16287 (2024) | Eleven commercial speech recognition services on university lectures | Accuracy varied widely between vendors and between samples | The abstract gives no WER figures and no native vs non-native results. |
Two results stand out. First, the best systems in the 2025 McGuire paper scored close together on read speech. The authors say those results approach human-level accuracy. Second, the best spontaneous result, 0.063, is higher than the best read result, 0.054. Real talk is harder than reading a sentence aloud.
Read the numbers carefully. McGuire reports Match Error Rate, not word error rate, so the figures do not line up directly with the WER numbers on vendor pages. The paper also has no native-speaker baseline. You cannot measure the gap between native and non-native speakers from it.
The paper also found that longer processing time did not reliably mean better accuracy. Speed and accuracy are separate questions. Do not pick a tool because it finishes fast.
Why accuracy changes from one speaker to the next
Your accent is one variable. It is not the only one, and you can control several of the others.
Hesitations, restarts and self-corrections
McGuire grouped disfluencies into fillers, repetitions and revisions. The systems handled each type differently. If you say "um" often, or restart a sentence halfway through, expect more errors.
The practical fix is simple. Finish the clause, then correct it. Do not stop and start again in the middle of a phrase.
Background noise and mic distance
A laptop mic across a loud room picks up more of the room than of your voice. Dictation then has less clean signal to work with. A headset mic near your mouth usually helps more than any setting change. We compare the options in our guide to the best microphone for Mac dictation.
Names and job-specific terms
Product names, client names and technical words cause errors for almost everyone. Plan for them. Keep a short list of the words that keep coming out wrong, and check whether your tool lets you add them.
Your language setting
If the wrong language is selected, the output is wrong no matter how clearly you speak. Check the language before you blame the model. We cover that and other common failures in Mac dictation not working and how to fix it.
How many people this affects
Ethnologue 2025, as cited by Rosetta Stone, estimates about 1.53 billion English speakers. About 1.14 billion of them speak English as a second language. That is roughly three-quarters.
Ethnologue's own page was not available to us, so that figure rests on Rosetta Stone's citation. It is still the best public estimate we found. If you speak English as a second language, you are the majority of English speakers. Tools should work for you. You still need to check that they do.
Accent is not the only voice factor that moves error rates. A 2020 PNAS study on racial disparities found an average word error rate of 35% for Black speakers and 19% for white speakers across five systems. The PNAS page would not open for us, so we rely on TechXplore's summary. That study is about race and dialect, not second-language speech. It still shows that who is speaking can change the result a lot.
Set up Dictation for English on your Mac
Apple's Dictation help page covers the basics. Work through these in order. Each one removes a common cause of bad results.
Start with Apple menu > System Settings > Keyboard. Scroll down if you need to, open the Dictation section and switch it on. If macOS asks, click Enable.
Next, set your languages. Click Edit next to Languages and add English, plus any second language you use. Apple says you can switch between languages while you dictate, using the language indicator next to the cursor or the Globe key.
Then set a shortcut you can press without thinking. Open the Shortcut pop-up menu, choose Customize and press the keys you want. Apple's example is Option-Z. Apple's page does not name a default shortcut, so check what your Mac uses before you test.
Check where processing happens. In the same Keyboard settings, confirm that general dictation is processed on your device and not sent to Siri servers. Optional sharing of audio recordings is a separate choice, so decide on it separately.
Finish with a test. Dictate two sentences into Notes, then press Escape to stop. If the text looks wrong, fix the language, the mic or the vocabulary before you change anything else.
Apple also says Dictation is not available in all languages or regions. Apple points to its macOS Feature Availability page for specifics. Check that page if your main language is not English.
Two languages on one Mac
Many people switch between English and a first language all day. Apple lets you set more than one dictation language and switch while you speak. Apple's page does not say whether one sentence can mix both languages. Assume it cannot until you test it.
Here is a worked example. You write client emails in English and personal notes in Spanish. Add both languages. Before you dictate an email, check the language indicator at the cursor. Dictate the whole sentence in one language, then switch for the next one.
Switching in the middle of a sentence is the case Apple leaves open. Test it with your own sentences before you rely on it for anything you will send.
Check the Apple Intelligence enhancement before you count on it
Apple's Dictation page says that with Apple Intelligence in macOS 27, dictation uses an on-device model to improve spelling, punctuation and capitalization. Apple says this is available in English on Mac models with M3 or later and at least 12GB of unified memory.
Apple's Apple Intelligence help page says the feature is not available in all languages or regions. It does not mention Dictation. The Dictation page is the one that describes this enhancement.
To check your Mac, open Apple menu > About This Mac. Look at the chip and the memory. If your chip is older than M3, or your memory is under 12GB, the enhancement does not apply on Apple's stated terms. In that case, plan for standard dictation behavior and run the test in the next section before you trust it.
Speaking habits that help dictation
Small changes to how you speak often help more than a new setting. Try these for a week and compare.
- Speak in full phrases, not one word at a time. "Send the report to the design team" works better than three separate bursts.
- Pause at the end of a thought, not in the middle of a phrase. A pause under 30 seconds does not stop Dictation.
- Say punctuation out loud when you need it, such as "comma", "period" or "new paragraph".
- Keep the same distance from the mic each time. A change of distance can change the result more than a change of accent.
Mistakes that make dictation look worse than it is
Most bad first impressions come from a few habits. Watch for these.
Judging accuracy from one sentence. One short test sentence with a name in it tells you very little. Use a paragraph or two.
Pausing for more than 30 seconds. Apple says Dictation stops by itself after 30 seconds without speech. You may not notice the stop until you look at the text. Keep thinking pauses short, or start a new session with your shortcut.
Fixing each word while you speak. Stopping to correct a word breaks the sentence. Finish the thought, then fix the text.
Skipping the read-through. Dictated text is a draft. Read it once, slowly, before you send it. Most dictation errors are easy to spot when you read them aloud.
Expecting names to work on the first try. If your tool has a custom word list, add names once and check again later.
How to test accuracy on your own voice
A short test beats any published number for your situation. Here is a method you can repeat.
- Pick a paragraph of about 120 words that looks like your real work. Client emails and project updates work well.
- Open the app where you normally write, such as Notes, Pages or Mail.
- Read the paragraph aloud at your normal speaking pace, using Dictation.
- Count the errors. A substituted word, a missing word and an added word each count as one error.
- Divide the errors by the total words, then multiply by 100.
Worked example: 120 words with 9 errors gives 9 divided by 120, which is 0.075. That is a 7.5% word error rate.
Run the same paragraph three ways. Use your normal pace, then a slower pace, then a quieter spot or a headset mic. Compare the three numbers. The condition that changes the most tells you what to fix first.
Your score is not directly comparable with McGuire's MER. It is comparable with your own results over time, which is what you need. For a guided version, the dictation accuracy checker on this site is a place to start.
Where Lazur fits
Lazur is a voice assistant for Mac. Dictation is one feature, not the whole product. Hold fn to dictate clean, formatted text at the cursor in any app. The app also keeps snippets, a personal dictionary and clipboard slots. The personal dictionary is the feature to use for names and terms that keep coming out wrong.
Dictation is unlimited on both paid plans. The app runs on Apple silicon and Intel Macs. Windows is planned, not available. Lazur listens and looks at the screen only while you hold the shortcut. Check pricing for current plan details.
We do not publish an accuracy number for Lazur's dictation. Run the 120-word test above on your own voice before you choose any app, ours included. If you want to try it, download Lazur for Mac.
Next steps
- Check your chip, your memory and your dictation languages against the notes above.
- Set your language and shortcut under System Settings > Keyboard > Dictation.
- Run the 120-word test in the app you use most, and write the error rate down.
- If the test fails because Dictation will not start, see Mac dictation not working and how to fix it. To compare Apple's tool with other options on price and privacy, read Apple Dictation alternatives for Mac.
- If you write academic text and language anxiety slows you down, read non-native English writing anxiety on Mac.
FAQ
Is Mac dictation accurate enough for non-native English speakers?
For many speakers, yes, but the published data is limited. The best 2025 study we found on non-native speech tested Whisper and AssemblyAI, not Apple's built-in Dictation. Run a 120-word test in the app you use for work before you rely on it for important text.
Does the on-device dictation enhancement work in every language?
No. Apple's Dictation page says the macOS 27 enhancement for spelling, punctuation and capitalization is available in English on Mac models with M3 or later and at least 12GB of unified memory.
Can I dictate in two languages on one Mac?
Yes. Add both in System Settings > Keyboard > Dictation, then switch with the language indicator next to the cursor or the Globe key. Apple's page does not say whether mixing both languages in one sentence works, so test that before you rely on it.
Why does Dictation stop when I pause?
Apple says Dictation stops automatically when it detects no speech for 30 seconds. Shorter pauses do not stop it. Press Escape, or your Dictation shortcut, to stop it on purpose.
Sources
- Dictate messages and documents on Mac (Apple Support) · Apple · checked 2026-11-07
- Use Apple Intelligence to communicate with ease on Mac (Apple Support) · Apple · checked 2026-11-07
- Automatic Speech Recognition for Non-Native English: Accuracy and Disfluency Handling (arXiv 2503.06924) · arXiv (Michael McGuire) · checked 2026-11-07
- Measuring the Accuracy of Automatic Speech Recognition Solutions (arXiv 2408.16287) · arXiv (Kuhn et al.) · checked 2026-11-07
- A Detailed Look at How Many People Speak English Globally (Rosetta Stone) · Rosetta Stone · checked 2026-11-07
- Automated speech recognition less accurate for blacks: study · TechXplore · checked 2026-11-07


