Binaural sound localization based on time-delay compensation and spatial grid matching

Binaural source localization is a popular technique in various applications, such as hearing aids, mobile robot, video conference, etc. However, robust binaural cues estimate and suitable localization strategy are always limiting its performance. In this paper, a new algorithm for binaural sound source localization is presented, which is a two-layer model. In the first layer, a spectral weighting generalized cross correlation-phase transformation (GCC-PHAT) method is presented for robust time-delay estimation, by which the probabilistic azimuths of sound source are obtained. In the second layer, an improved algorithm is introduced, which is named Compensated Interaural Intensity Difference (CIID). Based on the probabilistic azimuth localization results and CIID features, spatial grid matching (SGM) is presented to provide a Bayesian model for localizing azimuth and elevation. Compared with three other algorithms, experimental results show that the proposed method has a robust result even in noisy environments.