AI-scored speaking assessment
Six-criterion AI scoring for spoken language, with automatic anomaly detection and mandatory examiner review.
Six criteria, two sources of truth
Pronunciation and fluency come from an acoustic model; grammar, vocabulary, coherence and task achievement are banded by AI against the rubric.
The examiner always has the last word
Every score separates AI Score → Examiner Score → Final Score, and every change is kept forever.
Original recordings are kept
Raw audio and the full scoring history cannot be edited or deleted, so review and appeals always have evidence.