Reconstruction of Words From Brain Waves Feasible

In order to identify what acoustic information in speech sounds can be reconstructed from neural activity, Brian N. Pasley, Ph.D., of the University of California Berkeley, and colleagues utilized intracranial recordings from the nonprimary auditory cortex in the human STG.

Based on the data from auditory spectrograms, the researchers were able to show that slow and intermediate temporal fluctuations were accurately reconstructed using a linear model. A nonlinear sound representation based on temporal modulation energy was required for reconstruction of fast temporal fluctuations. The accuracy of reconstruction was highest within the range of spectro-temporal fluctuations that have been reported as necessary for speech intelligibility. During single trial presentations, readout and identification of individual words directly from brain activity was possible using decoded speech representations.

"This is huge for patients who have damage to their speech mechanisms because of a stroke or Lou Gehrig's disease and can't speak," a coauthor said in a statement. "If you could eventually reconstruct imagined conversations from brain activity, thousands of people could benefit."

Abstract
Full Text