Voiceitt learns atypical speech for conversations, dictation and meeting captions

Voiceitt is designed for people whose speech conventional recognition systems find difficult to understand. It learns from recordings of an individual, then converts their speech into text or a computer-generated voice. The company documents a browser-based service that requires an internet connection and currently focuses on English.[1]

Teaching the recogniser your speech

Onboarding starts with recording 50 phrases. The service then prepares a personalised voice profile, with further recording milestones allowing it to learn more about the speaker. Voiceitt’s help pages describe progressively larger sets, reaching 1,000 recordings; those levels measure submitted training material, not a clinical assessment of someone’s speech.[2]

A useful distinction is between learning a person’s sound patterns and learning the words important to them. Personal Vocabulary adds material such as names and frequently used terms. Shortcut Phrases serve a different purpose: a short spoken expression can trigger a longer message that the user has chosen in advance.[2]

The initial recording stage and the later personalised profile are separate stages. A person can explore the interface before that profile is ready. The sensible moment to assess recognition is after the relevant update, using language they actually need for conversation or work.[2]

Voiceitt’s FAQ says recognition varies with the speaker. Its recogniser supports spontaneous speech beyond the phrases used for training. The useful test is how faithfully the resulting text or audio conveys the person’s intended words.[1]

Making a video meeting readable

The Zoom integration carries the personalised recogniser into a meeting. The speaker needs an active Voiceitt account and completed onboarding recordings. Other participants can view the Voiceitt captions as guests, without creating Voiceitt accounts themselves. Their ordinary speech is handled separately by Zoom’s own captions.[3]

That shared viewing step is easy to overlook. Recognition on the speaker’s device becomes a communication aid when the people listening can also see the words, in the same conversation and at the right moment.[3]

Users helped change the design

A 2024 Assistive Technology paper described a development project that recruited 66 people with dysarthric speech. Participants selected uses relevant to their own lives and tested the app over three months or more, with technical support and interviews adapted to their needs. The work examined user involvement in product development.[4]

Their feedback led to faster voice training, simpler navigation, game-style features and switch access as an alternative to touch controls. These changes address a practical reality: the people who need speech assistance may also have difficulty operating the screen used to set it up.[4]

A separate University of Pittsburgh pilot is examining communication and daily use more directly. Its September 2026 registry update listed recruitment ongoing, with an estimated 20 participants and completion in December 2027. Planned measures include speech intelligibility, communication effectiveness, satisfaction and frustration. Those are planned evaluations, with no results posted in the record checked for this article.[5]

The design work points towards a useful definition of success: a person can express the message they choose, operate the relevant controls and make themselves understood in the activity they wanted to join. Recognition accuracy matters because it supports that complete act of communication.[4]

Report an error or suggest a correction