Related Experiment Video
Updated: Oct 28, 2025
Utilizing Repetitive Transcranial Magnetic Stimulation to Improve Language Function in Stroke Patients with Chronic Non-fluent Aphasia
Published on: July 2, 2013
Neuroprosthesis for Decoding Speech in a Paralyzed Person with Anarthria
David A Moses1, Sean L Metzger1, Jessie R Liu1
1From the Department of Neurological Surgery (D.A.M., S.L.M., J.R.L., G.K.A., J.G.M., P.F.S., J.C., M.E.D., E.F.C.), the Weill Institute for Neuroscience (D.A.M., S.L.M., J.R.L., G.K.A., J.G.M., P.F.S., J.C., K.G., E.F.C.), and the Departments of Rehabilitation Services (P.M.L.) and Neurology (G.M.A., A.T.-C., K.G.), University of California, San Francisco (UCSF), San Francisco, and the Graduate Program in Bioengineering, University of California, Berkeley-UCSF, Berkeley (S.L.M., J.R.L., E.F.C.).
Researchers developed a brain-computer interface to decode speech directly from brain activity in a paralyzed individual. This technology achieved real-time sentence decoding, offering a potential breakthrough for assisted communication.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Artificial Intelligence
Background:
- Restoring communication for paralyzed individuals significantly enhances autonomy and quality of life.
- Current assisted communication methods can be improved upon by directly decoding speech from brain activity.
Purpose of the Study:
- To investigate the feasibility of decoding words and sentences directly from cerebral cortical activity in a patient with anarthria.
- To advance assisted communication technology through brain-computer interfaces.
Main Methods:
- Implanted a high-density, multielectrode array over the speech-controlling sensorimotor cortex of a patient with anarthria and spastic quadriparesis.
- Recorded 22 hours of cortical activity during attempted speech of 50 words over 48 sessions.
- Utilized deep-learning and natural-language models to detect, classify, and decode words and sentences from cortical patterns.
Main Results:
- Achieved real-time sentence decoding at a median rate of 15.2 words per minute with a 25.6% word error rate.
- Post hoc analysis detected 98% of attempted individual words.
- Classified words with 47.1% accuracy using stable cortical signals over 81 weeks.
Conclusions:
- Successfully decoded words and sentences directly from cortical activity in a patient with anarthria and spastic quadriparesis.
- Demonstrated the potential of deep-learning and natural-language models for real-time speech decoding from brain signals.
- This approach represents a significant advancement in brain-computer interfaces for communication restoration.
