Pronunciation
quiz.

A minimal pair is two words that differ in exactly one sound and in nothing else. Ship and sheep are a minimal pair: same consonants, same syllable count, same stress, and one vowel — short /ɪ/ against long /iː/ — carrying the entire difference in meaning. So are thin and tin (/θ/ against /t/), right and light (/r/ against /l/), and cot and caught (/ɒ/ against /ɔː/).

The pair is the smallest possible experiment in pronunciation. Because everything else is held constant, a mistake can only come from one place. If you say sheep when you meant ship, there is no ambiguity about which sound went wrong — and no way to blame the consonants, the rhythm or the context.

That is also why linguists use them. A pair like pin and bin proves that /p/ and /b/ are separate phonemes in English rather than two ways of saying one sound, because swapping them changes the word. Languages differ in which contrasts they treat this way, which is exactly why some pairs feel effortless to you and others feel like the same word twice.

The minimal pairs guide works through the contrasts by sound family, and the vowel minimal pairs and consonant minimal pairs lists let you play both words of each pair back to back.

Auditory discrimination is the ability to hear that two sounds are different. It sounds trivial, and for the sounds of your first language it is — you have been sorting them since infancy. The difficulty is that a language's sound categories are learned, and by adulthood they are so automatic that sounds falling outside them get filed under the nearest category you already have.

That filing is why accent work stalls. A Spanish speaker whose vowel system has one high front vowel where English has two will hear both ship and sheep as the same vowel. A Japanese speaker whose language has one liquid consonant will hear right and light as the same word. Nothing is wrong with anyone's hearing: the sounds are arriving intact and being sorted into a system that has no separate box for them.

The practical consequence is an ordering. You cannot reliably produce a contrast you cannot reliably perceive, because you have no way to tell whether your own attempt landed. Practice without perception is practice without feedback — you repeat the word, it sounds right to you, and it is still wrong. Perception first is not a preference; it is what makes the practice self-correcting.

This is why the quiz on this page is a listening test rather than a speaking test. Getting eight or nine out of ten on a contrast means your ear has built the category. That is the point at which imitating the sound starts to pay off, and it is also the point at which the accent pages — American, British and Australian — stop sounding like variations on one accent and start sounding like three.

English has around twenty vowel sounds written with five vowel letters, so the spelling will not tell you what to listen for. Two contrasts do most of the damage for most learners.

Ship and sheep: /ɪ/ against /iː/

These differ in more than length. The vowel of ship is short and relaxed, made with the tongue slightly lower and the mouth a little more open; the vowel of sheep is long and tense, with the tongue high and forward and the lips spread as if you were beginning a smile. If you lengthen the short vowel without changing its quality you get neither sound. Listen for the tension, not the duration: bit and beat, fill and feel, live and leave. The long vs short vowels list has the full set.

Bat and bet: /æ/ against /e/

Both are short front vowels, which is what makes them hard. Bat uses /æ/, made with the jaw noticeably dropped and the tongue low — it is the more open of the two, and in American English it is often longer and slightly diphthongised. Bet uses /e/, made with the jaw higher and the tongue in a mid position. Watch a speaker's jaw and you can often see the difference before you hear it. Then try man and men, sad and said, bad and bed.

A routine that works

Pick one contrast and stay with it. Play both words of the pair back to back several times and do nothing else — no repeating, no reading, just listening, until the two stop sounding interchangeable. Then take the listening quiz above in Minimal pairs mode, which plays one word of a pair at random and asks which it was. Guessing at chance means the category is not built yet; go back to paired listening. Once you are scoring above 80% consistently, start producing the pair yourself and record it, because at that point your ear is a reliable enough judge to check your own output. Ten focused minutes a day on one contrast beats an hour spread across ten.

For the mechanics behind the sounds, read English vowel sounds and reading IPA. For the contrasts that separate the accents rather than the words, see American vs British pronunciation and the trap–bath split list.

Nothing in the bank is generated by guesswork. The 1448 IPA questions and the 372 difficult-word questions are built from this site's IPA dictionary, so the answer key is the same transcription shown on the word pages. The 20 minimal pair questions come from the same pair list the minimal pairs guide uses. The 92 spelling questions and 64 vocabulary questions come from hand-checked British/American tables; forms that are genuinely acceptable in both varieties — the -ise and -ize endings, for instance — are left out entirely, because a question with two right answers is worse than no question.

The two listening modes are not pre-written at all. They are assembled when you press start, from the voices your device and this site can actually speak, so the quiz never asks you to identify an accent it cannot play.