Overview
- In pilot results published in Nature Communications on January 19, the device recorded a 4.2% word error rate and a 2.9% sentence error rate among five stroke patients with dysarthria.
- Participants reported a 55% increase in satisfaction compared with basic word-by-word output, indicating more natural, usable communication.
- The washable choker captures throat vibrations and heart-rate patterns, then uses two AI agents—including an embedded language model—to reconstruct words and optionally expand brief phrases into full sentences via a nod command.
- Researchers note early-stage limits, including a small sample, a constrained vocabulary, and per-patient tuning, and say broader robustness and generalisability remain to be tested.
- The team plans a clinical study in Cambridge this year for native English speakers and says the approach could eventually aid conditions such as Parkinson’s disease and motor neurone disease.