Spatial unmasking for nearby speech sources in a simulated anechoic environment

Abstract

The ‘‘cocktail party effect’’ is traditionally studied with sources varying in azimuth, where binaural and level cues are large. In these conditions, speech intelligibility improves dramatically when the target speech and maskers are spatially separated. In the present study, we investigated spatial unmasking of nearby sources for several configurations of target and sources varying in distance and azimuth. Targets were sentences from the IEEE corpus and maskers were speech-shaped noise. Locations were simulated over headphones using anechoic head-related transfer functions (both individually measured and using a spherical-head model). Speech reception thresholds were measured adaptively, varying target level while keeping masker level constant. For nearby sources, overall energy effects (due to changes in target and masker level with distance) are large. Therefore, target level was normalized to equate the signal-to-noise ratio at the better ear in all conditions. This approach allowed direct measurement of binaural unmasking in the various conditions. Overall energy effects (removed by the normalization) are also reported. Results are compared with detection threshold measurements from similar experiments. [Work supported in part by AFOSR Grant F49620-98-1-0108 to BGSC and NIDCD Grant DC02696 to RYL.]