New AI-powered wearable helps stroke patients communicate by mouthing words

“This work establishes a portable, intuitive communication platform for patients with dysarthria with the potential to be applied broadly across different neurological conditions and in multi-language support systems,” the researchers write in Nature Communications.

The researchers tested the device on five stroke patients with the help of a 47-word vocabulary of common phrases. In this test, the system was able to decode 95.8 per cent of words and 97.1 per cent of complete sentences. The system also tagged speech with emotional cues to make communication feel natural. About 55 per cent of users expressed higher satisfaction compared to basic word-by-word output. 

How does the device work?

Revoice senses subtle vibrations in the throat and pulse at the carotid artery. When a user silently mouths the words, the sensors printed on the soft and flexible fabric band record the tiny movements. Following this, a small circuit transmits data to AI that interprets speech and emotional cues. The signals from the device are reportedly processed by two AI agents; while one reconstructs words from bits of silently mouthed speech, the other interprets emotional state and contextual information. 

While earlier devices pushed users to take pauses between words, Revoice analysed speech continuously in rapid 100-millisecond segments, leading to an experience that is much closer to natural conversation. According to the makers, the fabric sensors are highly sensitive and are able to respond to deformations as tiny as 0.1 per cent. The device has been built to be worn all day and is powered by a battery that can last a full day. 

Testing of the device showed high accuracy for stroke survivors. Initially, the team taught the system patterns from 10 healthy volunteers and later fine-tuned it for each patient. The study found that the performance hiked significantly around 25 repetitions per word, demonstrating that the system could adapt faster to individual speech patterns impacted by stroke. 

The results of the research, published in Nature Communications, can be beneficial to support people even with conditions such as Parkinson’s and motor neurone disease. Reportedly the researchers are currently planning a clinical study in Cambridge for native English-speaking dysarthria patients to study the viability of the system. 

It needs to be noted that the research is at an early, experimental stage, with very limited testing and no broad clinical deployment. The test was with only five patients, and the device still needs much more clinical study before it could be widely available.