Novelty detection of covariance among stimulus attributes in auditory perception

Abstract

Novelty detection is characterized by enhanced response to a stimulus with some property changed relative to expected input. Many reports examine sensitivity to deviations in physical acoustic dimensions, patterns, or simple rules, but fail to consider information in higher-order statistical relationships between dimensions. Here we report novelty detection that depends upon encoding of experienced covariance between complex acoustic dimensions (attack/decay, spectral shape.) Here, novelty is defined as violation of experienced covariance between otherwise independent acoustic attributes. Listeners primarily discriminated sound pairs in which attributes supported robust covariance (15 pairs, Consistent condition) and rarely discriminated sounds that violated this redundancy (1 pair, Orthogonal condition) in randomized AXB trials without feedback. Probability of occurrence for Orthogonal trials was minimized by withholding them until the final testing block. Discrimination accuracy for Orthogonal sounds exceeded that for Consistent sounds as well as that for control stimuli absent experienced redundancy between attributes. Increasing Orthogonal trial probability reduces this enhancement, as does acoustic similarity between Consistent and withheld Orthogonal sound pairs. Results parallel novelty detection as measured by stimulus-specific adaptation and mismatch negativity. Implications for high-level auditory perception and organization will be discussed. [Supported by NIDCD.]