Sound source localization method based on LDA classifier
Yue Yang, Xiaoyu Gu, Xinwang Wan · 2016
Methods based on pattern recognition such as support vector machine or neural network, can overcome the deficiency of traditional sound source localization algorithms which have an insufficient robust ability in the harsh environment of small SNR and severe reverberation. Among those methods, Naive-Bayes classifier has a high accuracy to locate sound source with a small amount of calculation and strong robustness. This paper presents a method that use linear discriminant analysis (LDA) classifier to locate sound source with generalized cross correlation phase transform (GCC-PHAT) function as the feature vector. Simulation results have proved that in harsh environment, LDA classifier locating accuracy is higher than Naive-Bayes classifier by 2%, especially in severe reverberation environment.