Pattern Analysis Based Models of Masking by Spatially Separated Sound Sources

摘要

Research is described in three areas: masked detection, sound localization, andneural network models of sound localization. Work on masked detection indicates that substantial reductions in masking of 8 to 18 dB can be realized when the signal is spatially separated from the masker in the free-field. This reduction in masking appears to be mediated by high-frequency information. Headphone-based studies of reproducible noise masking question traditional models of binaural masking, by showing unexpected relations between responses under monaural and binaural conditions. A new response technique has been developed to support work on sound localization. Neural network models of sound localization based on binaural stimulus cues can produce responses comparable to those of human observers. Our efforts in laboratory development and in planning the Conference on Binaural and Spatial Hearing are also briefly described.