Research by David A. Moses and colleagues studies the concept of Neuroprosthesis for decoding speech in a paralyzed person with anarthria. Those who are paralyzed may benefit from technology that enables them to communicate again. It's conceivable that a system that decodes words and phrases directly from such people's cerebral cortex activity would be a step forward. In patient’s whose spastic quadriparesis and anarthria were caused by a brain stem stroke, the authors recorded 22 hours of brain activity over 48 sessions as a participant attempted to pronounce words from a vocabulary list of 50. Deep-learning algorithms identified and classified words by analyzing brain activity patterns. These computational models and a natural-language model were employed in his research to compute next-word probabilities based on words that came before them in a sequence. With a 25.6 percent word error rate, words were decoded from brain activity in real time (Moses et al., 2021).
|
|
RSS Feed
