Binaural Cepstrum Coefficient and Its Application to Ground Target Recognition

Stereausis is a biologically motivated model proposed by Shamma which encodes both binaural and spectral information in a unified framework to simulate the processing of human binaural auditory system. In this paper, a new type of cepstrum coefficient is proposed based on this model. Two-channel acoustic signals are first processed by the stereausis binaural model to synthesize the spectral information and reduce the interference of noise signal. The binaural cepstrum coefficient is then extracted based on the diagonal vector of the stereausis model's output pattern, and is applied as feature to the multi-class acoustic target recognition. Learning Vector Quantization (LVQ) algorithm is implemented as the classifier and is tested by samples of vehicle acoustic signals. Experimental results show that binaural cepstrum coefficient improves both the performance and generalization of the classifier, especially at low SNR.