基于自适应有限差分粒子滤波的说话人跟踪

Speaker tracking based on adaptive finite-difference particle filteration

针对非线性说话人跟踪系统,提出一种基于自适应有限差分粒子滤波算法的麦克风阵列声源定位与跟踪方法.该方法在改进的粒子滤波框架内,采用适应性较强的布朗运动模型,通过计算麦克风阵列波束形成器的输出能量来构建似然函数,有效降低观测误差的不确定性对说话人位置估计的影响,一定程度上提升了说话人跟踪系统的精度.实验结果表明,该方法在基于麦克风阵列的说话人跟踪系统中具有较高的精确性.

On the basis of nonlinear speaker tracking system,a sound source localization and tracking method was presented based on adaptive finite-difference particle filteration（AF-DPF） of microphone array.Confined within an improved framework of particle filtering,Brown motion model with stronger adaptability was used in this method and a likelihood function was constructed by means of evaluating the output energy from beam-former of microphone array,so that the influence of uncertainty of observation error on estimation of speaker＇s position would well be reduced and the accuracy of speaker＇s tracking system enhanced to a certain extent.Experiment result indicated that this method exhibited a higher accuracy in the microphone array-based speaker tracking system.