Researchers Develop Brain-Computer Interface That Turns Inner Thoughts To Speech
The device works by translating brain activity into text and synthetic speech. In this case, the participant, 45-year-old Casey Harrell, had four microelectrode arrays implanted into the speech region of his brain. Just 25 days after surgery during the very first trial, the device achieved 99.6% accuracy with a limited 50-word vocabulary, requiring only 30 minutes of calibration data while Harrell attempted to speak.

On the second day of use, researchers expanded the system to a massive 125,000-word vocabulary. With only 1.4 additional hours of training data, the BCI reached 90.2% accuracy. With further refinements and more data, the system stabilized at an impressive 97.5% accuracy, sustaining that performance for more than eight months of continued use. During this time, Harrell used the BCI in natural conversations, averaging 32 words per minute across more than 248 hours of communication.
What sets this work apart is the rapid calibration and sustained reliability. Earlier BCI systems often required lengthy training periods and produced frequent word errors, making real-world communication difficult. By contrast, this new approach not only decodes words with accuracy rivaling commercial solutions, but also outputs them in a synthetic voice modeled on Harrell’s own pre-ALS recordings, giving him back a deeply personal sense of identity.
The study, published in the New England Journal of Medicine under the title "An Accurate and Rapidly Calibrating Speech Neuroprosthesis," represents a milestone in the field of assistive neurotechnology. While still in the investigational stage, the results suggest that speech restoration for people with paralysis may be closer than ever before.
For Harrell, the impact is profound. "Not being able to communicate is so frustrating and demoralizing. It is like you are trapped," he said. "Something like this technology will help people back into life and society." Check out the video above to see the device in action, or head to UC Davis' blog post if you want more details.