The Sound Source Localization Based On Microphone Array

Source localization based on microphone array can be used for a videoconference system to guide its camera to pick up the faces of the speakers automatically. However, background noise and room reverberations dramatically degrade the performance of most sound source localization systems. In addition to high accuracy, the location estimator must be updated frequently in order to be useful in practical real-time tracking and beamforming application. Aiming at these difficulties, mainly two methods for speaker localization using microphone array are discussed in this paper: delay and sum beamforming, time delay estimation and geometry localization; also a method based on Neural Net is introduced. Then a filter and sum beamforming method is presented to efficiently suppress the noise and reverberation. Also a ML parametric localization method is presented and compared with delay and sum beamforming method. As for localization method, a linear-correction least-squares approach is proposed to improve the accuracy of SI method, and LI method is also introduced.What is more, The modified CSP algorithm is proposed in order to improve the performance of CSP method. Higher-order statistics method is introduced to time delay estimation. A time delay estimation method based on onset signal and CSP is proposed, it's robust to noise and reverberation.In the end, a practical system for microphone array speaker localization, which is robust to reverberation and noise, easy to be realized in real-time DSP system due to its low computation cost and simplicity, are presented in the paper. Experiments were conducted in a normal room with high reverberation. The results conform all these performances, show that the new algorithm outperforms other algorithm such as beamforming, LMS adaptive filter or CSP method only.