- i24NEWS
- INNOV'NATION
- Israeli startup uses AI to give a voice to people with speech difficulties
Israeli startup uses AI to give a voice to people with speech difficulties
"The only thing preventing people with speech difficulties from participating in society is the communication barrier- Barvox is designed to eliminate that barrier"

For millions of people around the world, speaking is only part of the challenge. Being understood can be even harder.
An estimated 30 million people with autism are semi-verbal worldwide, alongside hundreds of millions of others living with conditions such as ALS, aphasia and other speech disorders that affect their ability to communicate.
An Israeli startup believes artificial intelligence can help bridge that gap.
Barvox, founded by entrepreneur Dror Levy, has developed what he describes as a personalized version of Google Translate—one that doesn't translate between languages, but instead learns an individual's unique speech patterns and converts them into clear, natural language in real time.
"If Google Translate translates French to English or Chinese, Barvox translates non-standard verbal expressions into natural language," Levy said. "Google learns languages. Barvox learns people."
The technology was inspired by Levy's 27-year-old son, Bar, who is autistic and semi-verbal.
For years, Levy served as his son's personal interpreter, receiving frequent phone calls from caregivers, teachers and coaches who struggled to understand what Bar was trying to say.
"Whenever he went to an activity, they would call me every hour asking, 'He asked for this or that—we don't understand. Can you translate?'" Levy recalled.
Levy demonstrated how the app works using a common interaction at home. When Bar says "shippui"—a word that normally refers to a barbecue in Hebrew—the application recognizes that, in his personal vocabulary, it means he wants lunch, allowing caregivers to respond immediately.
Without that personalized translation, Levy says, his son could be left misunderstood, frustrated and even hungry.
Unlike conventional translation software, Barvox creates an individualized language model for each user, learning how that specific person communicates rather than relying on standard speech patterns.
Once trained, the app listens to the user's speech and instantly translates their unique verbal expressions into language that others can easily understand, allowing conversations to flow naturally.
The goal is to remove one of the biggest barriers facing people with communication challenges: being understood in everyday life.
Speech therapist Orly Ziv says the technology could represent a significant advance over traditional augmentative and alternative communication (AAC) devices.
Many existing communication tools require users to scroll through symbols, words or images—or type messages manually—which can interrupt conversations and lead to frustration.
"Barvox can translate immediately what the person says—their own unique words—right away," Ziv explained. "It's much more natural, and conversations don't slow down."
She added that some users become so frustrated with conventional communication devices that they abandon them altogether.
"I've seen many iPads broken and thrown across the room because it's so difficult for people to find the words they want to communicate," she said. "This technology could make communication much easier."
Levy says the company's ambition extends far beyond his own family's experience.

Barvox aims to help an estimated 350 million people worldwide living with conditions that affect speech, including autism spectrum disorder, ALS and aphasia.
"Our success will be measured by whether we can help millions of people participate fully in society," Levy said. "They can work, have careers, build families and live independently. The only thing preventing many of them from doing that is the communication barrier—and Barvox is designed to eliminate that barrier."
As artificial intelligence continues to transform healthcare and assistive technology, Barvox is hoping its personalized approach will help ensure that more voices—no matter how they sound—can finally be understood.
