Voiceitt is speech recognition for atypical speech – built on a proprietary database of non-standard speech patterns rather than the general voice data every other engine trains on. It ships as an app, a Chrome integration and an API.
The gap it fills is stark. Mainstream speech recognition works well for speech close to its training distribution and fails almost completely outside it, which shuts out people with cerebral palsy, ALS, Parkinson’s, Down syndrome or the after-effects of a stroke – exactly the population for whom hands-free control and dictation would be most useful. A general model getting better at average speech does not help them at all; only training on atypical speech does.
A per-user training step adapts the system to an individual speaker, which is how the accuracy is reached. The API matters as much as the app: it lets other voice products become accessible rather than requiring people to use a separate one. Expect setup effort, since personalised training is the mechanism, not an optional extra.
