Medical AI and Assistive Tech

Researchers at the University of Cambridge have built and tested a soft, choker‑style wearable called Revoice that uses throat‑mounted textile strain sensors and dual AI models to turn silently mouthed words from stroke survivors with dysarthria into fluent, natural‑sounding speech in real time. The device, powered by a small 1,800 mWh battery for all‑day use, reads tiny muscle vibrations in the neck plus pulse signals that help infer emotional tone and context, then uses one AI agent to decode the words and another lightweight language model to expand them into complete sentences. In early trials with five stroke patients who had significant speech impairment, Revoice achieved a word error rate of 4.2% and a sentence error rate of 2.9%, for example converting a patient’s mouthed phrase "We go hospital" into a full, grammatically correct sentence. Unlike brain implants or eye‑tracking systems, the fabric choker is non‑invasive and designed to be washable and comfortable, and researchers say its low‑power, on‑device AI avoids the lags that plagued earlier silent‑speech systems. If future studies in larger, more diverse patient groups confirm these results, clinicians and disability advocates say the technology could significantly change post‑stroke rehabilitation and assistive communication options in the U.S., where dysarthria burdens hundreds of thousands of survivors.