Neural Networks Could Predict Speech
Researchers from Russia's HSE University and the Moscow State University of Medicine and Dentistry have developed a machine learning model that can predict words about to be uttered by a subject based on neural activity recorded with electrodes.
The technology, they say, can help the millions of people worldwide affected by speech disorders limiting their ability to communicate.
These devices, called speech neuroprostheses, are brain-computer interfaces programmed to predict speech based on brain activity. Unlike other systems that require large electrodes to be implanted in patients' brains, the system developed through this project is able to read brain activity from a small set of electrodes implanted in a limited cortical area.
For the experiment, the subjects were asked to read aloud six sentences, each presented randomly between 30 and 60 times. As the subjects were reading, the electrodes registered their brain activity.This data was then aligned with the audio signals and then fed into a machine learning model with a neural network-based architecture that was asked to predict the next uttered word based on the neural activity preceding its utterance.
The neural network achieved 55 percent and 70 percent accuracy for the two electrode configurations used in the tests.
"The use of such interfaces involves minimal risks for the patient. If everything works out, it could be possible to decode imaginary speech from neural activity recorded by a small number of minimally invasive electrodes implanted in an outpatient setting with local anaesthesia," said Alexey Ossadtchi, director of the Center for Bioelectric Interfaces at the HSE Institute for Cognitive Neuroscience and author of the report, in a statement.