A new soft, washable collar called “Revoice” uses ultra-sensitive sensors and AI to turn a few mouthed words into full, expressive sentences for stroke survivors. Researchers say it could restore independence without invasive brain implants.
For many stroke survivors, the hardest part of recovery is not knowing what to say, but being unable to say it.
A new wearable device called Revoice aims to change that by turning a few silently mouthed words into full, fluent sentences in real time — no keyboard, screen or brain implant required.
Developed by researchers led by Luigi Occhipinti, a research professor of smart electronics, biosystems and AI at the University of Cambridge, the soft, flexible collar is designed to help people with post-stroke speech impairment communicate more naturally and independently. In an early trial, it allowed stroke patients to express themselves far more clearly and reported a large boost in satisfaction with their ability to communicate.
The device targets dysarthria, a common condition after stroke that weakens the muscles of the face, mouth and vocal cords. People with dysarthria often know exactly what they want to say, but their speech comes out slurred, slow or in short, broken phrases.
“When people have dysarthria following a stroke, it can be extremely frustrating for them, because they know exactly what they want to say, but physically struggle to say it, because the signals between their brain and their throat have been scrambled by the stroke,” Occhipinti said in a news release. “That frustration can be profound, not just for the patients, but for their caregivers and families as well.”
Today, most stroke patients with dysarthria work with speech therapists on repetitive drills, practicing words and short phrases over and over. Those exercises can help rebuild muscle control and clarity over months or even years. But they do not always translate into easy, everyday conversation.
“Patients can generally perform the repetitive drills after some practice, but they often struggle with open-ended questions and everyday conversation,” Occhipinti added. “And as many patients do recover most or all of their speech eventually, there is not a need for invasive brain implants, but there is a strong need for speech solutions that are more intuitive and portable.”
Revoice is designed to be exactly that kind of solution.
Worn like a soft choker, the device contains ultra-sensitive sensors that pick up tiny vibrations from the throat muscles, along with the wearer’s heart rate. Those signals are almost invisible to the naked eye and inaudible to others, but they carry rich information about what the person is trying to say and how they are feeling.
Inside the system, two artificial intelligence components work together. One AI agent reconstructs words from fragments of silently mouthed speech, using the throat vibrations as a guide. The second AI agent interprets emotional cues from heart rate and other contextual information — such as time of day or weather — to expand short phrases into complete, expressive sentences.
To keep the device practical, the team embedded a lightweight large language model, or LLM, that can run on minimal power. Unlike the massive AI models that live in the cloud, this smaller version is designed to work on or near the device itself, supporting real-time use.
In a small study conducted with colleagues in China, the researchers tested Revoice with five stroke patients with dysarthria and 10 healthy volunteers. Participants wore the collar and silently mouthed short phrases. With a simple double nod, they could choose to have the system expand those phrases into full sentences using the built-in language model.
One example shows how the device combines speech fragments, emotion and context. A user mouthed the phrase “We go hospital.” Based on throat vibrations, elevated heart rate and the fact that it was late at night, the system expanded it to the more natural sentence: “Even though it’s getting a bit late, I’m still feeling uncomfortable. Can we go to the hospital now?”
In tests with the stroke patients, the system achieved a low error rate when reconstructing words and sentences, suggesting that the AI could reliably interpret what users were trying to say. Participants also reported a 55% increase in satisfaction with their communication, an early sign that the technology could make a meaningful difference in daily life.
The findings are published in the journal Nature Communications.
Unlike many existing assistive speech tools, Revoice does not require users to type letter by letter, track their eye movements across a screen or undergo surgery for a brain implant. Instead, it aims to feel more like natural conversation: the user silently mouths a few words, the collar senses and interprets them, and the AI fills in the rest.
The device is also designed to be comfortable and practical. It is soft, flexible and washable, making it easier to wear throughout the day, not just in clinical settings.
Although the first trial was small, the results suggest that the approach could be promising not only for stroke survivors, but potentially for people with other conditions that affect speech, such as Parkinson’s disease or motor neuron disease. The Cambridge-led team is planning a larger clinical study with native English-speaking patients with dysarthria to test how well the system works in real-world settings and over longer periods of time.
Future versions of Revoice could add multilingual support, recognize a wider range of emotional states and operate as a fully self-contained device for everyday use, without needing to be tethered to external hardware.
For Occhipinti and his colleagues, the goal goes beyond technology.
“This is about giving people their independence back,” he said. “Communication is fundamental to dignity and recovery.”
Source: University of Cambridge