On June 8, 2021, Voiceitt, an Israeli company founded in 2012, launched a consumer app that uses speech recognition tuned to atypical speech. The app is aimed at people whose speech is hard for ordinary voice systems to understand - those affected by cerebral palsy, brain injury, stroke, Parkinson’s, ALS, Down syndrome and similar conditions. Mainstream voice assistants are trained on typical speech and tend to fail this population; Voiceitt’s approach is to adapt to each individual’s pronunciation, cadence, breathing pauses and non-verbal sounds.
The app works by having the user record phrases in a short training step - Voiceitt cited roughly five minutes per phrase - which builds a personalized model. From then on the app recognizes that person’s trained phrases in real time and translates the output into clear text or standard synthesized speech, so the user can be understood by caregivers and loved ones or control smart-home devices. The pitch, in the company’s words, was that everyone deserves to express themselves and be understood.
Voiceitt arrived alongside Google’s Project Euphonia and Project Relate as part of a small field tackling the same gap: speech technology that works for everyone except the people who could most use it. What these efforts share is the recognition that “non-standard” speech is highly individual, so the solution is personalization rather than a single universal model.
Why business readers should care: Voiceitt is a focused example of building AI for an underserved long tail, where the commercial and human value comes precisely from serving users that general-purpose systems were never trained to handle.