Act out a real conversation. We give you each line — say it
clearly and we'll help with the tricky words.
Your line — say it out loud
How that went
Let's talk!
Pick a scenario. Your voice is understood and scored by this
app's own speech models — only the words you say are ever sent to the AI
partner, never your recording.
Headphones recommended — otherwise the
coach may hear itself. Tap “End call” to go back to hold-to-talk.
Fluency report
Voice calibration
Scores are decided by comparing a learner to a reference
of clear pronunciation. Out of the box that reference is built from a
computer voice, which is unnaturally even — so the bar sits somewhere no
real person speaks. Recording a few real people fixes that for everyone.
Use Preview the difference to see exactly what changes before
applying anything.
You'll read 30 short sentences (about 5 minutes).
Children get their own, simpler set — picked automatically.
Any clear accent is welcome — the goal is understandable English, not
one particular accent. Quiet room, normal speaking voice.
🔒 Locked
Calibration changes how pronunciation is scored for every
learner, so it's password-protected. There are two passwords: a
recording one for whoever is reading the sentences, and a
full one for applying the result.
Who is recording?
Children are scored against a
separate reference, so a child's recordings only count towards it if
you say so here. This is fixed by the first recording.
Recordings can't be re-made —
keep a copy off the server.
Read this out loud:
Check the recordings
Optional, but worth a spot-check on your first speaker:
listen for background noise or an unclear reading. The automatic checks
catch silence, distortion and misread lines — they can't hear a dog
barking behind an otherwise fine take.