Auditory Cortical Representation and Its Classsification for Passive Sonar Singnals

This study presents an auditory cortical representation for passive sonar signal and use it to extract features to categorize an unknown signal as being of man-made or natural origin. A ridge partial least squares (RPLS) model is given to establish the decision criterion, and regression coefficients are used to search for salient regions of the auditory cortical representation related to the classification task. In order to verify the utility of this method, some perceptual features motivated by music timbre research are used as predicators, and it is shown that the recognition accuracy of such an auditory cortical representation is slightly higher. Some acoustical analysis based on such two methods draw a conclusion that passive sonar signals of natural origin tend to possess more rhythmic transients or stationary noise.