Abstract:

Two models of spectral cue processing were compared to determine how the auditory system might extract directional information from pinna filtering. Previous experiments [E. Macpherson, J. Acoust. Soc. Am. 99, 2515 (1996)] suggest that the processing is robust to irregularity in source spectra, making effective use of pinna cues without utilizing prior knowledge of the source. The two candidate models attempt to achieve this by different means. The first uses correlation to find the best match between eardrum spectra (or their derivatives) and a set of stored templates associated with particular directions. The second makes the match based on prominent local spectral features such as peaks and notches. The models' predictions were compared to listeners' responses in localization experiments involving free-field presentation of stimuli with a variety of nonflat spectra. Correlational matching of the spectral gradient gave the best agreement overall, while feature-matching proved unrealistically sensitive to spectral irregularities. Neither model consistently predicted specific mislocalizations or stereotypical front--back confusions. Despite serious shortcomings, the correlational model demonstrates that a simple, physiologically plausible process can exhibit the required general robustness. The results do not support the hypothesis that local features are the spectral cues for sound direction. [Work supported by NIDCD.]