Towards Neural Activity Driven Vocalization Prostheses

Vikash Gilja, University of California San Diego

Abstract

Neural prostheses hold the promise of restoring lost function for individuals with motor, speech, and language deficits due to injury and neurodegenerative disease, and to advance our understanding of how the brain controls complex behavior. Although limb-based motor prostheses are actively studied and have yielded increasingly high performance, speech, language, and communications prosthesis development, while promising, is more limited. We aim to accelerate speech and communications prosthesis development with an avian model that complements ongoing human studies.

A central aspect of this work is the collection and curation of large-scale datasets that include continuous measurement of vocalization behavior across hours and days along with simultaneous neural activity from 100s of neurons from multiple brain regions, together with information on behavior-defining context manipulations. These data enable development and validation of machine learning models that map neural activity to behavior. As a complement to data acquired in human studies, our avian studies facilitate continuous data collection across multiple brain regions in highly controlled settings for larger cohorts of subjects. Thus, data from our studies can enable rapid prototyping of novel neural prosthetic system designs and can be applied to examine the potential for existing designs to generalize across subjects and behavioral contexts.