AI collar Revoice allows some stroke patients to speak again

Researchers at the University of Cambridge have developed a wearable AI collar called Revoice that uses throat vibration sensors, pulse signals and an embedded large language model to reconstruct speech; a small trial of five patients with dysarthria reported a 2.9% sentence error rate.

Cambridge engineers have developed a wearable collar named Revoice that uses artificial intelligence to reconstruct natural-sounding speech from silently mouthed words. The device combines sensors that detect subtle throat vibrations and pulse signals with an embedded large language model to predict full sentences. Researchers say the approach aims to be a portable, non-invasive alternative to brain implants. The work was reported in Nature Communications and led by the University of Cambridge's Department of Engineering. Key facts: - The device is called Revoice and was developed by researchers at the University of Cambridge. - Sensors on the collar capture throat vibrations and pulse signals that help infer emotional cues. - An embedded large language model is used to predict complete sentences from fragmented or silent speech, and the technology is related to the same AI class used in popular chatbots such as ChatGPT. - A trial with five patients who have dysarthria after stroke produced a sentence error rate of 2.9%. - The authors say they hope to extend the approach to other neurological conditions, to add multilingual support, and that extensive clinical trials will be required before wider availability. Summary: The Revoice collar reconstructs spoken sentences by combining throat and pulse sensing with an AI language model, offering a non-invasive speech-assist option for people with dysarthria after stroke. Early results from a five-patient trial showed low sentence error rates, and researchers plan further clinical testing and broader development of language and emotion decoding capabilities.