Duke University researchers have developed a brain implant that can provide communication based solely on thoughts. The device was designed to assist people who have language impairments or are unable to communicate verbally for any reason. Initial experiments showed good development prospects.

Experiments to convert brain activity into text and voice communication by scanning patients’ brain signals now allow “thoughts” to be translated into words at a rate of up to 78 words per minute. According to the study’s authors, this is like listening to an audiobook at half speed. Usually a person speaks up to 160 words per minute, which makes communication lively and natural. People with speech disorders also need more accurate sensors of brain activity to participate in this type of communication.

A group of scientists from Duke University and the University’s Biomedical Engineering Laboratory have created a brain activity sensor containing 256 sensors on a postage stamp-sized piece of plastic. The new sensor can capture signals from single neurons, allowing their activity to be determined with high accuracy.

Scientists would not be able to read minds directly. But they expected to determine with high accuracy the unspoken thoughts of patients (speech, whose signals are controlled by up to 100 muscles), based on a set of signals to the muscles of the speech apparatus (tongue, larynx and facial muscles). must be monitored). Therefore, the sentence spoken mentally had to be converted into signals to the muscles and everything the patient would say had to be reproduced according to this data read directly from the brain. In a patient whose speech apparatus was damaged, the thoughts would still remain in the cerebral cortex and the signals would not pass further, but would be counted by the sensor and the computer would provide the opportunity to speak.

An experiment with four patients showed that the average accuracy of mental word recognition was 40%, with a maximum of 84%. The recognition algorithm was trained in a “listen and repeat” manner. The patient spoke meaningless short combinations of letters, from which the algorithm learned to recognize brain activity in this or that combination of sounds.

Despite the relatively low percentage of voice recognition, the team of scientists speaks of success. The truth is that the algorithm was trained for only 90 seconds during a 15-minute test. The experimenters had the same amount of time as each patient. This occurred during planned operations on patients’ brains. After the neurosurgeons finished the surgery, they gave the scientists 15 minutes on their own schedule to work with the patients. Without access to the open brain, work could not be done in the specific area of ​​the cortex where the sensor was directly placed.

In the next stage, scientists will create wireless sensors that will work with patients under normal conditions, not in the operating room. One day, this will lead to useful brain implants that will convert thoughts into speech or digital messages.