AI-powered brain implant restores speech in paralysis patient after 18 years

Yahoo

4 days ago

Eighteen years after a brainstem stroke left her with near-total paralysis, Ann Johnson heard her voice again, thanks to a brain-computer interface (BCI) that decodes speech directly from brain activity.
Johnson, then 30, was a high school teacher and coach in Saskatchewan, Canada, when the 2005 stroke caused locked-in syndrome, a rare condition in which a person remains conscious but unable to speak or move.
Since then, she has communicated using an eye-tracking system at just 14 words per minute, a far from natural conversational speed of about 160 words per minute.
In 2022, she became the third participant in a clinical trial led by researchers at the University of California, Berkeley, and UC San Francisco aimed at restoring speech for people with severe paralysis.
The team used a neuroprosthesis that records signals from the speech motor cortex, bypassing damaged neural pathways to produce audible words.
Turning thought into voice
The device relies on an implant placed over the brain's speech production area. When Johnson attempts to speak, the implant detects neural activity and sends the signals to a connected computer.
An AI decoder then translates these signals into text, speech, or facial animation on a digital avatar.
Originally, the system used sequence-to-sequence AI models that required an entire sentence before producing output, creating an eight-second delay. In March 2025, the team reported in Nature Neuroscience that they had switched to a streaming architecture, allowing near-real-time translation with just a one-second delay.
To personalize the experience, researchers recreated Johnson's voice from a recording of her 2004 wedding speech.
She also selected an avatar to match her appearance, which can mimic facial expressions such as smiling or frowning.
Engineering for everyday use
Lead researchers Assistant Professor of Electrical Engineering and Computer Sciences Gopala Anumanchipalli (UC Berkeley), neurosurgeon Edward Chang (UCSF), and Berkeley Ph.D. student Kaylo Littlejohn say the goal is to make neuroprostheses 'plug-and-play,' turning them from experimental systems into standard clinical tools.
Future improvements could include wireless implants, eliminating the need for direct computer connections, and photorealistic avatars for more natural interactions.
The team envisions digital 'clones' that replicate not just a user's voice but also their conversational style and visual cues.
The breakthrough could help a relatively small but highly vulnerable population, including people who lose the ability to speak due to stroke, ALS, or injury, reclaim faster, more natural communication.
Researchers emphasize that the system only works when the participant intentionally tries to speak, preserving user agency and privacy.
https://www.youtube.com/watch?v=iTZ2N-HJbwA&t=1s
For Johnson, the trial was life-changing. 'I want patients to see me and to know their lives are not over now,' she said in a statement to UCSF. She hopes to one day work as a counselor in a rehabilitation center, using a neuroprosthesis to talk with clients.
With latency down to about a second and ongoing advances in AI modeling, the researchers believe practical, real-time speech restoration could arrive within just a few years, reshaping how technology gives voice to those who have lost their own.
Solve the daily Crossword