A system that detects the activity in the brain of a man with Sla translates his intentions into real -time sentences. And it allows you to modulate the sounds.
A man unable to articulate understandable sounds due to a neurodegenerative disease now manages to speak expressively, a check the intonation of words And even to reproduce very simple melodies. Thanks to a brain-computer interface that translates its vocal intentions, minutes but not onlyalmost instantaneously. The prosthetic voice system just described on Nature It is the closest thing to the natural speech ever achieved by this type of technologies.
How the brain system works
Casey Harrell is a 47 -year -old climatic activist suffering from Amiotrophic lateral sclerosis (SLA)a progressive neurodegenerative disease characterized by loss of motor neurons, The nerve cells of brain and spinal cord that allow the movements of the voluntary muscles. The pathology has weakened the muscles that Harrell used to speak: even if the man manages to produce sounds and move his mouth, the words that articulate are confused and not very intellectable.
The patient already had a series of 256 electrodes implanted for a previous study in the motor cortexa region of the brain that controls the movement. Maitreyee Wairagkar, neuroscientist of the University of California, Davis, and colleagues, have instructed an artificial intelligence system for decode the electrical activity in the motor cortex of man every 10 milliseconds. By transforming the motor commands for words into understandable sounds, then read by a synthetic voice.
Interjections and demand points
The brain-computer interface uses an approach that the authors of the study define “completely without restrictions”: decoding in real time not the entire words and not even their subunitations, but every minimum, single sound that man intends to produce. Including words without meaning and interjections like “uuhmmmm“or”eeeeeh“, which we use to attack one sentence to another. But the system also manages to translate The emphasis that is sometimes intended to give to a single wordor the highest vocal intonation than when a question is asked. Elements that make the speech much more natural than you could hope for systems of this type. Harrell even managed to sing some simple melodies, of three or four notes.
The difference compared to the past
For decades it has been working on systems for restore the ability to speak In paralyzed patients. Today Machine Learning algorithms can be trained to connect neural activities to the words of a predetermined vocabulary, even very large, with a repertoire of tens of thousands of words. But the brain-computer interface used by Harrell, capable of also produce meaningless interjection with the sole purpose of adding expressiveness to speechhas shown that it can also translate sounds that exist from a pre -established vocabulary.
In addition, the progress in the algorithms of used and the important number of electrodes implanted in the brain of Harrell allow the interface to recreate the words thought by the patient With a delay of only 25 milliseconds: about the time necessary so that our voice is heard from our ears, 40 times less the delay accumulated by other systems.
Towards a more natural language
The close times and similar to natural language, and greater expressiveness make the new system more suitable for being used in dialogues that contemplate a “blow and response” and various interruptions. Unlike the methods used so far, rather comparable to deferred exchanges, such as voice messages on WhatsApp. In addition, the interface trained to grasp the intonations lends itself to being used in tonal languagesin which tone variations determine different meanings.
