The experiment involved people with epilepsy.
Scientists have long been able to register the relationship between the neural activity of the brain and speech. However, recognition of signals and stream them into words and expressions are too complex and suffer from inaccuracy when compared with the usual speech. Meanwhile, the transfer on the fly of nerve impulses of the brain to speech might help people with injuries of the limbs and vocal apparatus, and that communication and the ability to use electronics and information, reports the Chronicle.info with reference on 3dnews.
Recently on the website of the journal Nature Neuroscience published an article of group of scientists from the University of California in San Francisco, which describes how to use recursive neural networks and brain-implanted electrodes you can learn to quickly translate the neural signals in the human brain in understandable words and expressions.
Note, the implant 250 of electrodes in the brain is no laughing matter, but in this case, no one specifically did not implanted. The experiment involved four patients with epilepsy and the electrodes they had implanted to monitor seizures. It so happened that some of the electrodes were in the areas of the brain in which there is a choice of words, compilation of expressions, and is feedback to the brain areas perceive their own speech.
Subjects were asked mentally, and then aloud to say from 30 to 50 sentences with a limited set of words. At the same time they filmed the signals from implanted brain sensors. The obtained data were transferred to the neural network for training, and received an intermediate score was given for analysis, another neural network. It turned out that the new algorithm for recognition of signals of the brain reduces the likelihood of an erroneous determination of words of up to 3 %.
Moreover, when one of the subjects mentally uttered the expression out of the proposed set of words, the probability of erroneous definitions of words decreased by 30 %. This means that the system can be trained for transfer to other patients, which reduces the training time for each individual.
The secret of the success of their experience, scientists see that they approached the problem of translation of thoughts into words from the point of view of the translator. The system analyzes the context and thereby reduces the probability of error. Work at the level of recognition of sentences, not individual words, gives the opportunity to eliminate the adverse impact of an accent, incorrectly or indistinctly spoken words and other ambiguities in the speech that still hindered the stream of nerve signals to speech.
