Why does 'ship' come out as 'sheep,' and can an app actually tell?

It happens because /ɪ/ (the short vowel in "ship") and /iː/ (the long vowel in "sheep") are close neighbors in the mouth, and whether an app can tell them apart depends entirely on what kind of listening it's doing.

30-second version: /ɪ/ and /iː/ are made in almost the same place — high, front of the tongue — but /iː/ is longer, tenser, and the lips spread further. Many languages have only one vowel in that region, so the two English sounds collapse into one. A phoneme recognizer that measures the vowel itself can catch this; a transcription engine that predicts words from context often can't, because "the ship sailed" and "the sheep sailed" are both plausible sentences.

Why this particular swap is so common

English packs an unusually large number of vowels into a small part of the mouth, and the /ɪ/–/iː/ pair sits right in the crowded high-front region. Many languages — Spanish, Japanese, and several South Asian languages among them — have a single vowel roughly in that space, used for both. Learners with that background don't have a "wrong" vowel; they have a system that never needed to split /ɪ/ from /iː/ before, so the ear hasn't learned to listen for the difference yet, which makes producing it reliably even harder than hearing it.

The two sounds aren't just "short and long" versions of the same vowel — /iː/ is also tenser and more front, so it isn't fixed by simply holding /ɪ/ longer. That's a separate, common mistake: stretching the wrong vowel doesn't turn it into the right one.

How sayit tells them apart

sayit's phoneme layer doesn't ask "which word is more likely here" — it measures the acoustic evidence for each vowel independently, frame by frame, and reports a per-frame confidence for every candidate sound. When you say "ship," the target phones are /ʃ ɪ p/. If what actually came out of your mouth acoustically matches /iː/ more than /ɪ/, the alignment records a substitution at that slot: target /ɪ/, produced /iː/. That's a fundamentally different signal from a transcription model, which would just as happily write down "sheep" if the surrounding words made that reading plausible and never tell you a swap occurred at all.

Open the flagged word and you'll see both IPA strings side by side, plus a "sounded like" hint if the substitution happened to land on a real word — which, for this exact pair, it usually does.

What the fix looks like

Say /ɪ/ with a relaxed jaw and a shorter, looser vowel — think of it as a quick, slightly lazy sound. For /iː/, spread the lips more (closer to a smile), raise the tongue a touch higher, and hold it noticeably longer. The length difference is real, but length alone won't fix it if the tongue height and lip spread don't also change — that's the tense/lax distinction most textbooks skip.

Drill it directly with ship vs sheep, and check nearby pairs that use the same contrast: bit vs beat and fill vs feel. The /phonemes/short-i and /phonemes/long-e pages have the full articulation steps.

How do I know it's landing?

Watch the word-level verdict across a few takes of the same sentence rather than one recording. A vowel that's still merging with its neighbor tends to swing between good and check depending on the take — that instability is itself useful information, because a vowel that's actually fixed stays good even when you say the sentence fast. If the same substitution (target /ɪ/, produced /iː/, or the reverse) keeps showing up across unrelated words, that's the pattern worth a focused week, not a single word worth memorizing.

Try it

Record: "The ship's crew counted six sheep on the hill." Both vowels sit two words apart, so a merge is easy to hear on playback. Open sayit free — no install, no card — and check whether the two words got two different verdicts.