Bayesian reconstruction of sound localization cues from responses to random spectra.

The directionally sensitive acoustics of the pinnae enable humans to perceive the up-down and front-back direction of sound. This mechanism complements another, independent mechanism that derives sound-source azimuth from interaural difference cues. The pinnae effectively add direction-dependent spectral notches and peaks to the incoming sound, and it has been shown that such features are used to code sound direction in the median plane. However, it is still unclear which of the pinna-induced features play a role in sound localization. The present study presents a method for the reconstruction of the spatially relevant features in the spectral domain. Broadband sounds with random spectral shapes were presented in rapid succession as subjects made saccadic eye movements toward the perceived stimulus locations. The analysis, which is based on Bayesian statistics, indicates that specific spectral features could be associated with perceived spatial locations. Spectral features that were determined by this psychophysical method resemble the main characteristics of the pinna transfer functions obtained from acoustic measurements in the ear canal. Despite current experimental limitations, the approach may prove useful in the study of perceptually relevant spectral cues underlying human sound localization.