基于粒子滤波的声源方位跟踪算法  

SOUND SOURCE ORIENTATION TRACKING ALGORITHM BASED ON PARTICLE FILTERING

作　　者：殷侠[1] 蔡卫平[1] 徐健[1] 陆泽橼[2] 

机构地区：[1]九江职业技术学院电气工程学院,江西九江332007 [2]中国科学院合肥智能机械研究所,安徽合肥230031

出　　处：《计算机应用与软件》2012年第12期77-80,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(60971098);九江职业技术学院科技项目(2011-22)

摘　　要：针对混响环境中说话人方位跟踪问题,提出一种基于粒子滤波的声源方位跟踪算法。该算法根据说话人的运动特点,在Langevin方程的基础上构建声源方位的动态模型,采用相位变换加权的可控响应功率作为定位函数,运用粒子滤波对声源方位进行跟踪。使用典型会议室环境下小型麦克风阵列接收的真实数据来做实验。结果表明,该算法能有效地实现随机走动说话人的方位跟踪,并且在水平角和仰角方向的均方根误差均小于5°。Aiming at speaker＇ s orientation tracking issue in reverberant environment, a sound source orientation tracking algorithm based on particle filtering is presented in this paper. According to the moving characteristic of the speaker, the algorithm sets up based on the Langevin equations a dynamic model of sound source orientation, the steered response power with phase transform weight （SRP-PHAT） is used as the localisation function, and the particle filtering is employed to track the sound source orientation. The real acoustic data recorded in a typi- cal meeting room using a small-scale microphone array is used for tracking experiments. Results show that the proposed algorithm can realise effectively the orientation tracking of the randomly walking speaker, in which, the root mean square error of tracking in direction of the horizontal angle and the elevation are all less than 5°.

关 键 词：麦克风阵列 声源跟踪 方位 粒子滤波 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]