2015 年 19 巻 1 号 p. 15-25
In this paper, MUSIC (multiple signal classification) algorithm is applied for a sound source tracking using particle filters. MUSIC algorithm is a high-resolution sound source localization method but has high computational costs. Therefore, a reduction of computational cost is attempted by using the MUSIC spectrum function as a likelihood function of particle filters. When there are two sound sources, however, a particle intensification problem arises simultaneously. Therefore, a method to overcome this problem is also proposed. Several experimental results in a reverberant environment show that the proposed method can track two sound sources in real time.