Sound-source Segregation Algorithm Using Binaural Model

Human beings can extract information about not only azimuth angle but also elevation angle of sound-source by using both right and left ear in the environment that noise exists or not. Direction of sound-source can be estimated through this, and the consequent sound-source segregation can be carried out. Conventional method that carries out direction estimation and segregation of sound-source by using human binaural model is executed in frequency band by using phase difference and level difference between signals entering both ears. Among these, sound-source segregation has the problem that gain of segregation filter linearly changes according to azimuth angle difference between direction of sound-source intended to be segregated and estimated direction of soundsource. Therefore, this paper suggests new segregation filter that has nonlinear characteristics, which can improve linear gain change that is problem of conventional sound-source segregation method. And performance evaluation of the proposed method is conducted by comparison of frequency response, after carrying out sound-source segregation.