Key points are not available for this paper at this time.
We propose a sound source separation method to extract target sounds from multiple sounds observed at binaural microphones worn by a user. The proposed method estimates the respective sound source position from the slope and variance of the spectral phase difference of two signals observed at the binaural microphones. Based on the estimated sound source positions, the proposed method separates all the sounds and extracts only the sounds that arrived from the target area specified by the user. Simulation and real-environmental experiment results show that the proposed method efficiently gives the sounds whose sources exist in the target area.
Kawamura et al. (Thu,) studied this question.