Voiceitt Built Speech Recognition for Voices That AI Never Understood

Voice assistants, phone menus and dictation software are built on one quiet assumption: that everyone speaks the same way. For millions of people with cerebral palsy, Parkinson's disease, ALS, stroke or brain injury, that assumption fails. Their speech is often slurred, uneven or softly voiced, a pattern researchers call nonstandard speech or dysarthria, and mainstream speech recognition simply does not understand them. Voiceitt, a company founded in 2012 and headquartered in Tel Aviv with a subsidiary in Stamford, Connecticut, built the first commercial speech recognition system designed for exactly these voices, and it matters because voice is becoming the default way the world interacts with technology, from meetings to medical care.

The scale of the problem is larger than most people realize. According to the CDC, cerebral palsy affects roughly 1 in 345 children in the United States. The Parkinson's Foundation estimates that nearly one million Americans live with Parkinson's disease. The CDC also reports about 795,000 people in the United States have a stroke each year, and stroke is one of the most common causes of dysarthria. Many of these individuals retain the ability to speak, but their voices diverge enough from typical speech that commercial voice AI transcribes them inaccurately or not at all.

Why Ordinary Speech AI Fails

Speech recognition systems are trained on enormous datasets of typical voices. When a voice falls outside that distribution, word error rates climb dramatically. For someone with dysarthria, the result is a world of technology that effectively shuts them out: phone trees that cannot route their call, dictation that garbles their words, and conference calls where their contributions go untranscribed. The problem compounds as voice becomes a primary interface for work and daily life.

Voiceitt's founders saw this gap up close. Co-founder Sara Smolley has described how her family's experience with Parkinson's disease shaped the company's mission to make voice AI accessible to everyone, regardless of how they speak. Rather than treating nonstandard speech as an edge case, the company made it the entire focus.

A Model That Learns Your Voice

Voiceitt uses a hybrid approach. Each user trains the app by recording a short set of words or phrases, which creates a personalized model tuned to their specific speech patterns. That personal layer works together with a foundation model that Voiceitt has trained on large amounts of atypical speech data, collected with consent from the communities it serves. The system runs in real time, converting speech into text or voice output, and it keeps improving as a person uses it, mirroring the way human listeners adapt to a familiar voice over time.

The company also treats users as collaborators, not test subjects. A study published in 2024 in the journal of Communication Disorders recruited 66 participants with dysarthric speech to use the Voiceitt app and feed structured feedback directly back to the developers. That kind of participatory design loop is rare in commercial AI and it shows in the product.

From the Living Room to the Meeting Room

The most powerful thing about Voiceitt is not a standalone app but where the technology now lives. An integration with Amazon Alexa lets users with nonstandard speech control their lights, music and smart home devices by voice, the same way everyone else does. In 2023, Voiceitt and Cisco announced an integration that brings accessible captioning and transcription to Webex Meetings, so an employee with dysarthria can be understood on a video call with colleagues. Cisco Investments is also a Voiceitt backer.

Voiceitt also offers an API, allowing other software makers to build recognition for nonstandard speech into their own products. Recognition at ITU's AI for Good arena followed, including a first prize win at an AI for Good startup event in March 2024. Each integration matters because accessibility works best when it is embedded in the tools people already use, not siloed in a separate app.

Independence, Dignity and Employment

For users, the stakes are deeply personal. Being able to speak a command that a machine obeys restores a measure of independence for people who may otherwise depend on caregivers for small daily tasks. Being transcribed accurately in a work meeting determines whether someone with a speech disability can hold a job in an economy that runs on video calls. ITU's AI for Good program has highlighted Voiceitt precisely because accessible speech technology touches employment, education and independent living at once.

The company's broader bet is that inclusion is a design choice. Most voice AI was trained without nonstandard speakers in the room. Voiceitt proves that when people with dysarthria are included in the data, the design loop and the product roadmap, the technology learns to listen. It is a small correction with an enormous reach, and a reminder that the most important AI breakthroughs are sometimes not about bigger models but about whose voices get to be heard.