Researchers at the University of California, San Francisco разработали a neurointerface that enabled a paralyzed man to speak. To decode the electrical impulses from the cerebral cortex they used artificial intelligence.
According to the scientists, this is the world’s first technology capable of assembling sentences from whole words, rather than from individual letters or symbols.
The researchers implanted an electrode grid in the patient’s sensorimotor cortex, which controls speech production, to record impulses and process them with deep learning algorithms for further decoding.
They noted that the subject was able to compose short sentences with up to 93% accuracy and a rate of more than 18 words per minute.
The data used to train the algorithm were collected over 48 sessions, lasting 22 hours. The scientists recorded brain signals when the patient attempted to pronounce 50 flashing words on a screen. They then trained the model to detect and classify them according to patterns in the recorded cortical activity.
According to group leader Edward Chang, they achieved these results thanks to progress in artificial intelligence and decoding neuronal signals that were not available five years ago.
The team is currently working on expanding the vocabulary and improving speech speed. In the future they intend to broaden the study to include more participants with severe paralysis and communication impairments.
Earlier in May, Stanford researchers presented a neurointerface that allowed a paralyzed man to type text messages at a speed of 90 characters per minute, spelling out individual letters and punctuation marks.
In July, researchers from MIT demonstrated a robot that helps disabled people dress safely.
In March, researchers at the University of Georgia developed an AI backpack to help the blind navigate their surroundings.
Subscribe to ForkLog news on Telegram: ForkLog AI — all the news from the world of AI!