Free English pronunciation test

Intelligible English is mostly a matter of individual sounds. Listeners forgive an accent easily; what makes them ask you to repeat is a sound that has been replaced by a neighbour — "sink" for "think", "berry" for "very", "sheep" for "ship", "wery" for "very". A useful pronunciation test therefore has to score below the word level, sound by sound, and tell you which specific sounds are being swapped, dropped or blurred.

This test uses five words chosen so that, between them, they cover the sounds most learners struggle with: the two "th" sounds, the English "r" and "l", the short and long "i" vowels, "v" against "w" and "b", and consonant clusters at the end of a word (the "-rld" of "world"). Each word is scored as a sequence of sounds against a native reference, so a single weak sound shows up on its own line rather than dragging down the whole word unseen.

What it does not measure is your accent. A Spanish or Indian or French accent is not an error, and this test does not try to make you sound like a newsreader. It measures whether each sound you produce is the sound an English listener expects — which is what decides whether you are understood.

The engine aligns your recording to the expected sounds of the word, then rates how closely each sound you produced matches native productions of that sound. The scores you see (0–100) are per sound and per word; the word score is not an average that hides a problem, and the weakest sounds are ranked to the top so you can see what to work on first.

Read the numbers like this: 85 and above is a sound native listeners will not notice; 70–84 is clearly understood but audibly non-native — usually a vowel slightly off or a consonant a little weak; below 70 is a sound that is likely being heard as a different sound, and one such sound can undo an otherwise good sentence. Most learners get one or two sounds well below the rest. That is the useful result: those are the sounds to practise, and the rest can be left alone.

The scoring was checked against people, not just against itself: on a public research set of 2,500 recordings by 250 English learners, each rated by five human pronunciation experts, the engine’s sound-by-sound judgments closely track the human ratings. Two things still affect a single take — a noisy room and a very quiet microphone both lower scores — so if a result surprises you, record the word once more in a quiet place before you believe it.

People search for both, and they are the same recorder with a different question. A pronunciation test asks "which sounds are wrong?" — it is the diagnostic. An accent test asks "how far am I from a native speaker overall?" and summarises the same per-sound results as a single picture. If you want to know what to practise, start here; if you want a one-number read on how you come across, take the accent test.

A low score tells you which sound; fixing it needs three more things. First, hear the contrast until you can reliably tell the two sounds apart — most learners who say "sheep" for "ship" also cannot hear the difference yet, and no amount of repeating fixes what you cannot hear. Second, learn the physical move: for "th", the tongue tip touches the top teeth; for the English "r", the tongue never touches anything; for "v", the top teeth rest on the lower lip. Third, record the same word again and watch the score move — the feedback loop is what makes practice stick.

Every sound this test scores has its own page with the mouth position, the words it hides in, and audio to compare against. For longer material, the practice page scores any English text you paste — a sentence from your presentation, a line from a song — sound by sound, the same way.