ABSTRACT
This chapter focuses on possible mechanisms that underlie feature tuning or feature selection in human auditory recognition. It examines some experiments whose findings suggest that considerable flexibility exists in the feature extraction process. In all experiments, the features that listeners use in perceiving the timbre of complex sounds were investigated. If one were to adopt a property-list approach to the feature extraction stage for timbre perception, it would be necessary to specify a list of important timbre attributes. Feature extraction plays a fundamental role in many theoretical treatments of auditory pattern recognition. A strict property-list model would direct people to search for evidence regarding invariant auditory feature detectors, whereas a process-oriented model would have they look for common principles underlying feature extraction across a wide range of different stimuli and tasks. Weighting parameters for each feature were estimated for individual listeners by fitting the model to the observed confusion matrices using a standard gradient technique.
