音像仰角推定に関する両耳聴モデル改善の基礎検討

Abstract

In this study, the binaural model for the sound image elevation estimation was extended and experimentally verified under adverse perceptual conditions in a car cabin. First, the relationship between the spectrum of left and right ear input signals and sound image elevation in a simulation of a car cabin environment and in an actual car cabin environment was learned using a neural network. Then, cross validation of the proposed method was conducted under the conditions of a sound field where detection and discrimination of the spectral notches, an important cue for sound source elevation perception, is difficult, and as a result, high generalization performance was obtained. The model was then further extended and its accuracy was again tested in the interior environment of a car with noise. Results verified that the proposed approach exhibited the same trends as the conventional method.