Consonant discrimination in the inferior colliculus of young and aged rats
Complex acoustic stimuli are thoroughly encoded and processed along the primary auditory pathway to give reliable and relevant information about the environment, and elucidating the neural coding mechanisms is essential to informing clinical attempts to combat auditory dysfuntion. Receiving a uniquely diverse set of ascending and descending inputs, the inferior colliculus (IC) is a site of intricate temporal processing. In this work, natural and modified human speech is used to investigate discrimination of voice onset time (VOT) in the spiking output of IC neurons. A template-matching classification model is proposed in which single stimulus presentation responses are correlated with aggregate spiking trends. Single units are found, in general, to perform well above chance on a 4-VOT paradigm but are vulnerable to stochastic mistakes on a trial-by-trial basis. Integration of classifier outputs over a population of units is shown to reduce variability and increase performance on single stimulus presentations. An important goal of the proposed model is to free the classifier from an assumption of absolute time alignment between trials and templates. Although exact temporal alignment aids classification accuracy, populations of units are shown to perform even without such knowledge. As neural computations are further understood, it is important to monitor how these processes are impaired in a pathological state. Young adult animals are contrasted with aged animals to investigate age-related hearing loss. For VOT discrimination in the central nucleus of the IC, decline of performance with age is noted for stimuli which are spectrally more difficult for the rat auditory system, but this deficit is missing for stimuli carried on broadband noise which more strongly drives neural responses. This is consistent with the idea of hidden hearing loss in which impairments are not apparent in perceptually easy clinical diagnostic tasks. Additional analyses and extensions of the proposed methods are described. Evidence is found for categorization of the 4 VOT stimuli into two groups which supports human perceptual studies. Established spike distance and wavelet-information metrics are applied to these data and their potential use for similar studies is weighed. Ultimately, an ideal extension of the classifier to continuous time monitoring by all units and a downstream integrator is detailed to most realistically model ongoing perceptual processes.
Bartlett, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our