How we use AI

What is generated

The scenes you practise on are made by AI: the dialogue, the voices and the video. Every scene carries a mark saying so, inside the application, on the scene itself.

We say this plainly because it matters for what you are learning: the speech in a scene is modelled on how people speak, not recorded from a particular person.

What is not generated

The phrases people bring in are written by people. The topics, the levels and the order in which material is given are editorial decisions, not model output.

What we do with your voice

When you say a line out loud, the recording is sent to our servers, scored, and deleted on a schedule. It is never published, never shown to other people and never used to train a model that leaves our infrastructure.

The score you get back — the sounds, the pace, the completeness, the intonation — is computed by our own models running on our own machines.

Video people upload

If you upload a fragment of your own, it is checked automatically and, when doubtful, by a person. What it may contain and what happens to it is set out in the terms of the application.

Which models

Scene generation and dialogue construction use external providers; recognition, alignment and pronunciation scoring run on models we host ourselves. The list of providers changes as we measure them, and this page changes with it.

Who to ask

Questions about any of this go to contact@sayvibe.app. The company behind SayVibe is SayVibe Technologies LLC, Georgia, Tbilisi, Chugureti District, Lochini Street 3, floor 2, apartment 5.