The development of neurology and electronics has brought many improvements in the treatment of nerve damage. Researchers can already be proud of the achievements we could not have imagined even years ago, a partial vision restoration for the blind.
In a recent experiment, researchers from the University of California, San Francisco the brain implant partially restored the ability to talk to paralyzed patients and others who can't speak nerve damage.
So far, most speech aids have been based on moving the muscles in the face or moving the eyes. The lack of this process is slowness – the average has moved 10 words per minute, which is significantly below the average of 150 words per minute. To speed up the process, scientists used machine learning algorithms that transformed brain activity into words and phrases.
The speech synthesis process took place in several stages. University of California researchers are Data on volunteers with normal brain function have been collected for several years. Volunteers participated in experimental treatment of epilepsy, which included inserting electrodes directly into the brain, and researchers gathered data on brain activity in the speech centers, while volunteers read hundreds of sentences.
Next stage researchers set up a decoderthat converts acquired brain signals into instructions for moving virtualized dictionaries (including lips, tongue, jaw and throat). For decoders, they created a synthesizer that converts virtual movements into speech.
Previous speech research focused on direct translation of brain signals into speech, but recent research has shown that the main task of brain centers is to send teams to move the muscles and organs needed for speech. Brain waves themselves have very little effect on how words really sound.
Using the above methods, researchers successfully synthesized many words and phrases, although the results are incomprehensible. In the analysis of artificially synthesized speech, volunteers succeeded in finding out about 80% of the words expressed in 101 speeches.
The results are very encouraging and the next point for researchers is to use such devices in real time, as the process is still quite time consuming. The use of algorithms for machine learning is very useful because they have the opportunity to easily acquire the knowledge they have acquired.
Intro Photography: Unsplash