Abstract
Traditional sound source localization algorithms employed the TDOA(Time Difference of Arrival) between microphones in a microphone array to estimate the angle of incidence. Even though the computations of these methods are not huge, the spatial resolution is affected by the sampling rate. In this thesis we use a high resolution algorithm called multiple signal classification (MUSIC). In the multiple-source case, this method not only can determine the number of sources, but also their angles of incidence. Then we use particle localization method (PLM) to realize two dimension sound source localization. In practice, we notice that the computation load of MUSIC algorithm is too heavy to realize in real time. So we adopt several methods to solve this problem. In practice, the performance of the system is also affected by reverberation, and the steadiness of this method is not ideal. Thus, to improve the performance we use Particle Filter and Kalman Filter to track the sound source. From the experiment results, we can notice that the Kalman filter converged faster than the particle filter. The bias of the Kalman filter when it converged is also smaller than the particle filter. The ability of the Kalman filter to track moving source is also better than the particle filter.