Statistical modeling of natural auditory source event perception
Richard E. Pastore, Jesse D. Flint, Jeremy Gaston · The Journal of the Acoustical Society of America · 2007
Challenges to understanding auditory source perception can be understood by contrast with classic psychophysics. Psychophysics studies the relationship between properties of sound and perception; by mapping changes in a single perceptual dimension as a function of controlled, systematic variation in a single acoustic property. This severely limits the degrees of freedom for both the listeners strategy and the investigators interpretation of the complex, often nonlinear, sound-perception relationships. Source perception investigation not only adds the original event as an analysis stage, but also multiple levels of complexity to every component in the analysis stages. Natural source events are complex and, although acoustic properties must reflect source event structure, variation in many acoustic dimensions of the produced sounds is determined both separately and interactively by attributes that vary with all source properties, including the one(s) investigated. Furthermore, physical acoustic dimensions seldom map directly to perceptual dimensions. Finally, listeners have many degrees of freedom in assigning importance weighting to the multitude of sound attributes, differ in their weightings, and are seldom optimal. The current paper discusses statistical modeling approaches for developing a systematic understanding of the relationships among the source-sound-perception components to event perception. [Work supported by NSF grant BCS-0213666.]