Functional Organization of Speech Processing Areas and A Systematic Approach to the Cocktail Party Problem
- Author(s): Hullett, Patrick W.
- Advisor(s): Schreiner, Christoph E
- et al.
The brain is a physical system that can perform intelligent computations. We are interested in nature of those computations to understand how the brain does intelligent things. To that end we have focused on two particularly fruitful questions that were tractable given the current state of knowledge and resources: What is the organization of processing in human speech centers? And, how does the brain solve the cocktail party problem?
To address the first question, we recorded superior temporal gyrus activity in awake human subjects passively listening to speech stimuli using electrocorticography. The high spatial and temporal resolution of this recording technique combined with maximally informative dimension analysis made it possible to compute high density spectrotemporal receptive field maps in a region of the brain specialized for speech perception. Based on these maps, we found that human superior temporal gyrus has a strong modulotopic organization - a higher order analog of tonotopic organization that has not been previously identified in any human or non-human auditory area.
To investigate the mechanisms by which neural systems solve the cocktail party problem, we created animals that are specialists at extracting vocalization information in the face of by noise-rearing rats and testing them behaviorally to show specialization. Through single unit recordings from primary auditory cortex, we identified a subpopulation of neurons that can extract vocalization information in the face of noise. Although the prevalence of these neurons is the same in both groups of animals, neurons from specialized animals extract information at significantly higher rates. Further receptive field analysis will give insight to the underlying mechanism of this ability. This work demonstrates the ability to create animals specialized at solving the cocktail party problem and a method to identify neurons that contribute to this specialization. This approach can be applied to different classes of noise to generate and refine models of cocktail party processing.