Researchers from Stanford University (USA) have achieved record-breaking accuracy in recognizing human inner speech. Using technology of implanting electrodes in the motor cortex and artificial intelligence, they were able to “read” the thoughts of the experiment participants with an accuracy of 26 to 54%. This is the highest rate of inner speech recognition to date.
A similar system was previously used on a patient with amyotrophic lateral sclerosis who could not speak. The brain-computer technology allowed the patient to reproduce words through articulation, but this was very tiring. The new approach allowed participants to simply imagine the spoken words.
Electroencephalography recorded the activity of motor cortex neurons even during “mental speech.” Artificial intelligence matched these signals with phonemes and composed words and sentences from a dictionary of 125,000 units. The participants of the experiment could imagine entire sentences, and the computer system decoded them in real time.
The results of the study, published in the journal Cell , open new perspectives for studying the neural mechanisms of inner speech and may become the basis for technologies to assist people with speech disorders. However, the systems still need further development for practical application.
