Back to Tutorials
Real speech-delivery data, captured once, from your actual voice
Pronunciation is the one criterion measured from real audio data, not estimated from your transcript — captured once, during your Part 2 long-turn answer. Here’s exactly what’s measured and how it maps to your band.
Real data, not a guess — captured on Part 2
Microsoft Azure’s speech engine analyses your actual voice during Part 2’s long turn — “the richest single sample of your speech in this test.” It’s bonus diagnostic detail, not a separate score — your Pronunciation band already reflects your whole test.
5 real numbers, each out of 100
Accuracy, Fluency, Overall pronunciation, Prosody (stress and intonation), and Completeness — 5 real measurements from your own voice, each scored 0 to 100.
How those numbers map to your 1.0–9.0 band
85–100 → 9.0. 75–84 → 8.0–8.5. 65–74 → 6.5–7.5. 55–64 → 5.0–6.0. Below 55 → below 5.0. Continuous natural speech rarely scores a literal 100, so 85+ already means excellent, native-level delivery.
Scored completely on its own
Pronunciation is never influenced by your grammar, vocabulary, or how developed your answer was — a response can have real grammar mistakes and still score highly here, judged only on how clearly and naturally it sounded. Any word scoring below 50 is flagged individually as a “word to practise.”
DEB
4 things to know about pronunciation scoring