Accent guesser: what accent do I have?

You read a fixed passage of about 100 words. A Whisper-based AI model (Whisper is an open-source speech model) turns your voice into an acoustic fingerprint, and a classifier trained on public voice datasets and on recordings from TonePerfect learners, each labelled with the speaker’s accent or first language, compares it with the accents it knows. It first estimates how likely English is your first language, then gives its top guesses with calibrated percentages. American and Canadian English are so close that it names them together.

Native English: American English, Australian English, Canadian English, English (England), Irish English, New Zealand English, Scottish English, South African English. First languages: Filipino (Tagalog), German, Indian (Hindi / South Asian languages), Malay / Indonesian, Mandarin Chinese, Russian, Spanish, Turkish, Vietnamese, and, less reliably for now, Cantonese (Hong Kong), French, Japanese, Korean, Polish. Not yet: Arabic, Italian, Portuguese (Brazil), Ukrainian, and every language not listed here: the guesser does not recognise those speakers well enough yet. If your first language is not on the list, it guesses the closest accent it knows, which is often wrong. When you tell it your first language after a guess, that answer is what lets us add it.

Because it is a guess from 40 seconds of speech. The common reasons: your first language is one it was not trained on; you speak more than one language or have lived in several countries; your microphone or room changed how you sound; or two accents are simply close. Tell it what your first language is after the guess: those answers are how it gets better.

We do not publish an accuracy figure yet, because we have not measured one on real browser recordings. On research recordings of people reading a passage it is often right for the accents it knows, and the right answer is often in its top three, but it is clearly worse for some languages. We will publish per-language numbers once enough people have told us whether it was right.

No. It guesses which accent your English sounds closest to, among the accents it knows. An accent is not where you are from, what passport you hold or who you are, and many people sound like a place they have never lived. It is a game and a practice aid, not a test to make decisions about anyone.

It is analysed and then deleted: it is never written to disk, and it is held in memory for at most 15 minutes only so that the “Keep my recording” box can work (answering the question without ticking it deletes it at once). Nothing is kept unless you tick “Keep my recording to help improve the guesser” after your result; then it is stored anonymously, without your name or email, and you can ask us to delete it. The same recording is also scored for pronunciation by our English engine, which does not keep it either. No signup, no account.

Yes. It runs in your browser with no signup and no download, a few guesses a day. The TonePerfect app, on iPhone, Android and the web, adds daily practice for the sounds it finds.

Not quite. Many people search for an “accent detector” when they want exactly this: a guess of which accent they have. Our English accent test answers a different question, how close each of your sounds is to a standard American reference, and our AI accent detector names the sounds that give your accent away. Neither of those two guesses which accent you have.