Listen to a natural sentence, repeat it once, and let AI score your accuracy, intelligibility, and fluency, in seconds.
In the real campus scenarios the test actually uses, every repeat is corrected down to the sound, scored on ETS-aligned dimensions, and tracked weak-spot by weak-spot.
Every word you say is checked against the original, then each slip opens the exact sound that tripped you, and how to fix it.
She acknowledged the lack of evidence in her report.
Accuracy, intelligibility, and fluency roll into one band on the 1–6 scale ETS now uses, mapped to CEFR.
Missed lines come back as focused retries, and each issue is tracked attempt by attempt, from still-off to fixed.
Crisp, short vowel — /læk/. That’s the target.
Shorter this time — the vowel still drifts a touch long.
Vowel ran long — /leɪk/, not /læk/.
Practice the campus and community situations the test actually uses, from the library desk to the museum floor, several pulled straight from real exams.
Before you can repeat a word of it, you first have to catch it in real time and hold the whole clause in your head. There is no transcript scrolling alongside and no slow replay to lean on.
That listening-under-pressure is the part TOEFL speaking quietly tests, the part pronunciation drills and pre-scripted answers never touch. Repeating a sentence back is what forces you to train it: your ear and your working memory, not just your mouth.
Most speaking scores don’t break on the sounds. They break on what you didn’t catch.
Not a crowd, not a vibe — here's what every grade actually rests on.
Every sentence is pulled from an actual TOEFL recall — not prompts we made up.
A real speech engine checks each word down to the IPA, across three scored dimensions.
Each retry is logged Still → Improved → Fixed — and can only ever move forward.
You hear one sentence at natural native pace and say it back exactly. It's a 2026 TOEFL Speaking task, and it quietly tests three things at once — how well you caught the sentence, held it in memory, and delivered it clearly — not just pronunciation.
Each repeat is rated on accuracy, intelligibility, and fluency, then mapped to a single 1–6 band on the 2026 scale. As a working guide, 4.0 is 'good enough to apply' to most programs and 4.5–5.0 is competitive — and Listen & Repeat is the more forgiving speaking task, with about 75% of our learners' attempts reaching 4.0 or above.
Because it isn't only a pronunciation test. A repeat can break at three separate points — you mishear the sentence, lose it in working memory, or deliver it unclearly — and most people are strong at one and weak at another. That's also why repeating the same sentence ten times rarely helps; TOEFLAIR pinpoints which of the three is costing you so you drill the real cause.
Since the 2026 reform, ETS itself scores Speaking with AI first, and on the countable parts — delivery and language use — its automated engine agrees with human raters about as closely as two trained humans agree with each other. We built our grader on that same ETS standard and have run it across 674 real attempts; it is strongest on those measurable parts, and every score is broken down into the exact words or moments behind it, so you see the reasoning, not just a number.
Yes. You can do one full sentence right on this page and get real AI feedback with no account or credit card; sign up only when you want to save progress and keep practicing.
Start free on TOEFLAIR, one sentence at a time, with feedback you can act on.
Get started free