Step one: the phone listens
When you speak, a speech-recognition model running on the device turns your audio into text. There is no round trip to a server for this step, which is why the tutor can react quickly and why the basics keep working on a bad connection. The model was chosen for accented English: it was measured on speakers of many first languages, not only on native speakers, because an app for learners has to understand learners, not only native speakers.
The tutor's voice is also generated on the phone. Your personalised voice profile, if you make one, is created and kept on the phone and never uploaded.
Step two: what did the sounds do?
In Mimic practice you repeat a short phrase. Mimica lines up what you said against the target and looks for the shifts that recur: th becoming t or d, r and l swapping, v and w, the short i in "ship" stretching into "sheep", a final consonant disappearing, stress landing on the wrong syllable. These are the patterns a first language leaves on English, and they are the same ones listeners notice.
You see which sound moved, instead of a score out of a hundred. A number only tells you that something is wrong. A highlighted sound tells you what to fix next.
Step three: say it in words you understand
Each flagged sound comes with a short explanation of what changed and, where it helps, a hint in your own language: what the mouth does, what the closest sound in your first language is, and why the English one is different. The aim is that you can make the correct sound on purpose, not that you memorise a diagram.
Step four: hear it in your own voice
For English, Mimica can play the same phrase back with clear rhythm, vowels and stress, in your own timbre. This is the Voice Mirror. The reference is no longer a stranger. It is a clearer version of you, which is a target you can actually reach.
Which steps use the cloud
- On the phone: speech recognition, the tutor's voice, your voice profile.
- In the cloud: the tutor's conversation replies and the detailed pronunciation assessment. These go to contracted AI providers, and Axielo does not keep copies of your recordings on its servers.
The details are in why your voice stays on your phone.
What the feedback tells you
Mimica tells you which sounds in your speech are shifting, whether they shifted less than yesterday, and whether a listener would still have understood you. That is the part that changes how your English sounds to other people.