Scientists from UC Berkeley and UC San Francisco have created a new brain-computer interface with “naturalistic speech for people with severe paralysis." In an important study published in Nature Neuroscience, the development addresses a significant hurdle in speech neuroprostheses and is a huge step forward, allowing people who lost the ability to speak to communicate in real-time with one another.

Researchers say they employed artificial intelligence to create a streaming method that converts brain signals to audible speech in nearly real-time. The method parallels how existing voice assistants (e.g., Alexa, Siri, etc) rapidly decode speech, producing a “nearly-synchronous voice streaming” that sounds as natural and believable as human speech.

Cheol Jun Cho, the co-load author of study, with his team essentially used a neuroprosthesis that would sampled neural data from the motor cortex - a part of the brain that controls speech, and then use AI to decode these brain signals into speech. The researchers tested it on a 47-year old woman named Ann, who has been unable to speak for roughly 18 years.

For the clinical study, Ann was implanted with electrodes implanted on the surface of her brain, which recorded neural activity while she mentally attempted to speak of sentences on a monitor. The system then translated that activity in real time to audio speech, powered by an AI model that approximated her voice before her injury.While brain computer interfaces (BCIs) have existed for some time, the study was significant for its near real-time translation of a patient's thoughts into audio. Previous BCI systems were decades in development but were usually delayed by 8 seconds on average to convey a single spoken sentence. The new technology holds the possibility to improve quality of life for those with illness or injury, such as ALS or stroke induced paralysis.