Speak Level

Learners rarely have an objective, repeatable way to find out how good their speaking actually is — or what to work on next. Self-assessment is unreliable, formal exams are expensive and infrequent, and most apps drill vocabulary and grammar rather than measuring spontaneous speech.

speak-level takes a one-minute speaking sample and returns three things: an estimated CEFR level from A1 to C2, a breakdown by sub-skill — pronunciation, fluency, lexical resource, grammatical range — and recommendations ordered by impact, grounded in the sounds and words the speaker actually struggled with.

The pipeline transcribes and aligns the audio, scores each dimension independently, then maps the profile to a level. Pronunciation and fluency scoring run on Azure Speech; lexical scoring and recommendation generation run on Amazon Bedrock through the Amplify AI Kit. Results are stored, so progress is trackable over time rather than being a one-off snapshot.

It is built API-first: everything the web app shows is available as JSON, with per-key configuration and usage quotas, so other course platforms can submit audio and receive scores programmatically. It doubles as the free assessment companion to Back to Fluency — measure your level here, train it there.