Resolution of steady-state sounds in simulated auditory space

Abstract

A simulated acoustic equivalent of a sound source placed on the perimeter of a circle having a 4-m radius in the frontal horizontal half-plane was generated by obtaining, in 5-deg azimuthal steps, head-related transfer functions measured in both ears of an artificial head placed in an anechoic room. When an arbitrary sound is passed through the digital filter defined by the left and right transfer functions corresponding to a given angle, the sound will acquire a subjective azimuth comparable to the one at which the transfer function was measured [J. Blauert and P. Laws, Acustica 29, 273–277 (1973)]. This technique was used to measure the resolution of pairs of simultaneous steady-state stimuli (frequency-modulated, amplitude-modulated, and pure sinusoids or noises). The listener was required to identify the relative location of the two sounds. In one set of experiments, the position of the two sounds was fixed, and thresholds were obtained for the modulation- or carrier-frequency difference at which the two sounds could be localized. In other sets of experiments, the pair of sounds was constant and the minimum audible angular separation was determined at which association of each sound with a particular locus was possible. Generally, both kinds of resolution were better around the midline than at the sides. Angular resolution at the sides approached the threshold for the minimum audible angle between serially presented sounds only for pairs of complex (FM-) stimuli having little spectral overlap; for most other sound pairs, spatial resolution was possible only at angular separations of 60° and larger. Results suggest that segregation of the two component sounds and their correct localization constitute competing processes. Spatial resolution will be inefficient when either the localizability of the sounds is poor (e.g., for long-duration sine waves) or when the spectra of the two components display large regions of overlap (e.g., for two noise signals). The relationship between the data and the ‘‘cocktail-party effect’’ is discussed.