Improved speech source localization in reverberant environments based on correlation dimension

Abstract

Speech source location estimation in a noisy, reverberant environment has attracted much attention recently. It was found that the localization method through calculating the steered response power (SRP) is more robust than time-difference-of-arrival (TDOA)-based localization method. The method places equal emphasis on each microphone pair in calculation of the SRP. In a room, each microphone is usually differently affected by noise and reverberation. Thus the performance of the SRP-based localization method will degrade. In this paper, we place a weighting on each microphone pair according to their correlation dimensions to compute SRP. Simulations show that the proposed correlation-dimension-based source localization method is more robust to noise and reverberation than the SRP-PHAT (phase transform) algorithm