Only 6% of the errors were made when it recognized more than 1.1 000 words.
A neural interface has been developed by scientists from the United States that allows you to read paralyzed people’s thoughts and those with speech difficulties with remarkable accuracy. Researchers report that the installation is only about 6% wrong, which speeds up the process for voicing thoughts.
«Clinical trials have shown that a system that connects directly with paralyzed brains can be used to allow them to communicate meaningful sentences. The system recognized more than 1.1 000 words and made only 6% errors in our experiments. This allowed it to decode thoughts at a speed 29 characters per minute.According to the researchers, “
Scientists have developed a number of neural interfaces to transform the brain activity of patients into written or oral speech in recent years. A set of electrodes is implanted in the brain that allows paralyzed persons to control the mouse cursor, press the buttons, and draw letters on the computer screen. Paralytics can communicate freely with others around them thanks to this device.
An international team of neurophysiologists, led by Edward Chan (USA), has created a system that allows you directly to read the thoughts of speech-related patients. This method simplifies the typing process, allowing you to quickly decode entire sentences and words with minimal errors.
Professor Chan and his co-workers rely on the radiotelephone telephone alphabet for their work. This alphabet is used by international organisations to transmit the letters of Latin alphabet in the most unambiguous manner possible. Each character is encoded within the framework with a short English term, such as foxtrot (foxtrot) and uniform (uniform).
Scientists suggest that machine learning systems can recognize brain activity generated from thoughts about each word. Neuroscientists had this idea in mind and enlisted the assistance of a paralyzed volunteer, whose brain was implanted using electrodes. These conductors were used by the scientists to record the brain signals generated by the patient when he was pronouncing the words that encode the alphabet’s letters.
The scientists used the received set of impulses to train an AI network that was capable of recognizing code words in realtime and using them to create sentences and words. To improve the quality of this system, scientists have organized the AI system in such a way that it ignores all signals that do not match the built-in dictionary of 1.1 thousand of the most commonly used words in the English language.
This approach allowed the volunteer only to speak a few words per min and not make any errors. The total number of errors was less than 6%. Scientists hope that the system’s speed and accuracy will improve the quality of life for paralyzed and dumb patients in the future.