Researchers from UC Berkeley and UC San Francisco have used a brain-computer interface (BCI) to help people with severe paralysis speak naturally.
The researchers fixed a problem called latency. Latency is the delay between thinking and speaking. They used artificial intelligence (AI) to change brain signals into sound quickly enough to enable severely paralyzed patients to speak naturally.
The researchers have described the methods and results of this study in a paper published in Nature Neuroscience.
In a press release issued by UC Berkeley, the researchers explain that their method decodes brain data fast and makes speech sound smooth and natural. This helps people who lost speech. The researchers have been running a trial with special electrodes that read brain activity.
The method works with different brain tools. Microelectrode arrays, or MEAs, go into the brain. Non-invasive sensors, like sEMG, sit on the face. The system reads the motor cortex, which the part of the brain that controls speaking. AI turns those signals into words.
Testing and Improving Speech
The researchers tested it with a patient named Ann, who cannot speak. She thought of sentences like “Hey, how are you?” The system mapped her brain signals to those words. They used AI to make audio since Ann has no voice. They used her old voice to make the audio sound like her. Before, decoding took 8 seconds per sentence. Now, sound comes in 1 second. Ann can keep talking without stops. The fast system stays accurate.
They also tried rare words like “Alpha” and “Bravo.” The system learned them too. Ann said it feels more real. She hears her voice right away. The researchers want to add tone and excitement next. This could make speech even more natural. The researchers plan to keep keep improving it, and express confidence that this work sets up future advances for BCIs.