How can non-native speakers get accurate English voice typing?

Accurate English voice typing starts with the right settings and a fair test, not with the tool's headline score. Choose the English variant you speak, use a microphone close to your mouth, and dictate a passage you already know to see what each tool gets wrong. In a 2025 study of non-native English, Whisper reached a match error rate of 0.054 on read sentences ¹, but the same systems differed in how they handled filler words and self-corrections ². Your results will depend on how you speak, so measure them. The same steps apply whether a tool calls it speech to text, dictation or voice typing.

Choose the English variant you speak, from the list on your device

The English variants differ by platform, and the choice affects which model you get. Apple's Dictation list for the Mac includes 15 English language and regional variants, including English (India) ³. Apple's newest dictation model, the one in macOS 27, is available for only 9 English variants ⁴, so the variant you pick can decide which model you use.

Check the list for your device, then choose the variant closest to your speech, and test the second-closest if the first one scores poorly.

Read sentences are accurate, and free speech needs more checking

The research supports a plain rule. Read sentences give the best results, and spontaneous speech is harder. The 0.054 figure above comes from read sentences, and the author describes it as approaching human-level accuracy. Free speech is a different test. The same study found the systems differed in how they handled disfluencies, such as filler words, repetitions and revisions ².

In practice, that means your planned sentences will come out cleaner than the thinking you do aloud. If you pause to search for a word or restart a sentence, expect more of that text to need editing. Write or outline the key sentences first, then dictate them, and you will usually get more accurate text.

Speed gains are real, but they are a ceiling

Dictation is much faster than typing for English text in a laboratory. In one controlled study on an iPhone, English entered by speech reached 153 words per minute, against 52 on the keyboard ⁵. That figure counts the finished text, including time spent correcting errors, but it used short messages in a quiet room. Treat it as a ceiling for planned, short text, not a forecast for your email or your reports.

Checking a dictated draft takes longer in a second language

Reading rates are lower for readers with English as a second language, according to a meta-analysis by Brysbaert ⁶. Dictation saves the typing, but you still have to read the draft, and that step may take longer for you than for a first-language reader. This is our inference from that finding, not a result the study measured, so test it on your own work.

What to do: a short plan

1. Set the variant. Choose the English variant you speak and write, from the list on your device.
2. Use a headset. A close microphone can reduce the noise and echo that reach the recognizer. Apple says a headset microphone may help in a noisy room or a room with excessive echo ⁷.
3. Dictate in phrases. Plan each sentence, then say it at a steady pace. Pause at the end of a sentence rather than in the middle of a word.
4. Test on a passage. Use about 200 words of your own text, count the errors, and divide by the number of words. That is your word error rate. Compare two tools on the same passage.
5. Add your names and terms. If the tool supports custom words, add the names, places and technical terms that it gets wrong most often.
6. Type the sensitive parts. Names, figures and legal wording are the errors that cost the most. Check them, or type them yourself.

The question on whether dictation can help non-native speakers at work applies this plan to email. If your accent is the main problem, read why dictation does not understand your accent. To adapt a tool to your voice, see can you train dictation software to your accent. For the same steps in another language, the guide to dictation languages covers the choices for each platform.

Where the evidence is thin

The main study we found tested speakers from several first-language backgrounds on a limited number of recordings, with a match error rate rather than the word error rate most tools report. The published figures cannot be compared directly with other tests. Read them as a guide to where the errors come from, and use your own passage for the decision.

For the broader research, see our guide to speech recognition and accents. The languages overview lists every guide in this series.

Yap, which we make, is an on-device dictation app for Mac and Windows, so you can run the same passage through it and through the built-in tools on your device. For prices and downloads, see pricing and the download page.