POSTER DETAILS
Temporal continuity and learning auditory objects: implications for solving the “cocktail party problem” and auditory perception
Ronald W. Ditullio, Chetan K. Parthiban, Eugenio Piasini, Vijay Balasubramanian, Yale E. Cohen
Date / Location: Sunday, 10 July 2022 / S01-151
Add this poster event to your 🗓
Add this poster event to your 🗓
A fundamental problem animals face is the unsupervised/minimally supervised learning of objects. Auditory objects present special challenges because they superpose, as opposed to occlude, and because of the large variability within auditory categories. A famous example of the challenge of learning and discriminating auditory objects is the “cocktail party problem”', learning and discriminating vocalizations amid the clutter of many similar stimuli. The mechanisms by which the brain solves this problem are not fully elucidated. One solution suggested by efficient coding theory is that neural circuits exploit the statistical regularities of the environment to solve perceptual tasks. Temporal regularities are a defining feature of many ethologically relevant stimuli, such as human speech and animal vocalizations. Accordingly, we hypothesize that efficient learning and discrimination of auditory objects can be achieved by training circuits to extract temporal regularities features. It follows from this hypothesis that unsupervised learning of the temporal regularities in stimuli (e.g., vocalizations), can support learning to discriminate between different vocalizations. To this end, we applied Slow Feature Analysis (SFA) to extract temporally regular features, i.e., those that vary in time as slowly as possible, from macaque vocalizations. Consistent with our hypothesis, we found that a linear classifier could discriminate effectively between vocalizations projected onto only five of the features found by SFA. These results suggest that the ``slow'' temporal features of auditory stimuli may be sufficient for parsing auditory scenes, providing a powerful computational mechanism that the brain utilizes for auditory perception.