Progress in speech decoding from the electrocorticogram

Abstract

Recent advances in neuroimaging methods have improved our ability to explore the neurological processes underlying speech and language. As a result of these investigations, it is now possible to decode aspects of speech directly from neural activity toward the development of neuroprosthetic devices for individuals with severe neuromuscular and communication disorders. Much of what is known about the neural correlates of speech articulation and perception is based on lesion and cortical electrical stimulation studies, as well as modern non-invasive neuroimaging. Though extremely important to the current understanding of brain function, traditional neuroimaging methods are primarily limited by the spatial and temporal resolution of the imaging technique. Electrical activity measured from the cortex, or electrocorticography (ECoG), offers several advantages over other neuroimaging modalities for characterization and real-time decoding of brain activity. Specifically, ECoG is well-suited for the study of speech and language owing to its unique spatial and temporal resolution capabilities that allow it to accurately capture the fast-changing dynamics of the large cortical networks underlying speech processing. This review presents the current progress of ECoG-based speech characterization and decoding studies, including an overview of prior neuroimaging studies, ECoG representations of speech production and perception, and a discussion of future directions.