Free
On-device practice
- Record or upload; pace, fillers, pauses, pitch variation
- Transcript editing with filler highlighting
- Compare attempts of the same task
- Student build kit and workbook
Schools, clubs and coaches
Speaking practice with evidence, and student privacy by design.
Students practise a short talk and see numbers they can check. Teachers stay in control of every AI step. No student accounts, nothing stored on our server.
Plans
Classroom pilot 2026–27
AI steps, teacher-controlled
- Speech recognition with word timing
- Structure check with quotes verified against the transcript
- Demo code per class; per-device rate limits
- Session plan, host checklist, onboarding call
School / club licence
For a whole school or programme
- Sign-in through your own identity provider
- Your subdomain and branding
- Data processing terms for your district
- Teacher training and an annual learning review
Pilot terms and licence pricing are set with each partner. We don't sell or share student data, and we don't use it to train models.
Why teachers like it
- Observable, not judgemental. “6 filler words in 19 seconds” starts a better conversation than “you seemed nervous”.
- Honest about the machine. Students see when the recognizer dropped an “um” they actually said — a live lesson in checking AI.
- Fair comparisons. Only the same task, the same source and the same metric version are compared.
- Maths in the open. Rates, thresholds, differences — every number shows its formula.
What we measure, and what we never do
| We measure | We never |
|---|---|
| Words per minute | Score a student's “speaking ability” |
| Filler words (um, uh, you know…) | Infer emotion, confidence or personality |
| Pauses ≥ 0.6 s, two independent rulers | Rate accents |
| Pitch variation in semitones | Diagnose speech or language conditions |
| Opening / body / closing, with quotes | Keep recordings or transcripts |
Status, honestly
Live since November 2026. Engineering, security and accessibility checks pass on every release (unit tests, browser end-to-end tests, an accessibility scan, user-persona journeys). Not yet established: learning outcomes with real classes — that is what the pilot is for. We publish what we have and have not tested.