Automatic Recognition of Sound Direction in the Median Plane of Human Ears

Abstract

The human ear orientates the sound source in the horizontal direction base on the binaural effect, while the orientation of human ear in median plane angle is based on auricle effect. This paper is based on a research device, which measures the sounds from sound sources at different angles in the median plane. Through acquisition of sound data at intervals of 5° between 0° and 180° in the median plane, the spectrum envelope of the sound is extracted to form data samples. The separability of data samples is studied by using three kinds of Minkowski distances. A convolutional neural network is constructed to carry out the research of automatic recognition the median plane angle of the sound source. The research proved that by using Convolution Neural Network (CNN), we can precisely orientate the direction in the median plane with a higher precision.