Speaker localization based on ratio between the second and the first eigenvalue

Abstract

To locate the position of the speaker, there are many methods, among of them, the method using steered response power is very popular. By phase transform, the steered response method (SRP-PHAT) is robust; with the noise increasing, the performance of the method becomes bad. Using principal eigenvector, the steered response power (SRP-PE) method is better than SRP-PHAT. In this paper, we propose an improved SRP-PE method, which only uses the frequencies with small ratio between the second and the first eigenvalue to estimate the sound source location. Simulation results show that similar to the SRP-PE algorithm, the localization accuracy of the improved SRP-PE method is better than the SRP-PHAT method; moreover, compared to the SRP-PE, the proposed algorithm has lower computational complexity.