基于sigma点H∞滤波的说话人跟踪方法

Sigma Point H_∞ Filtering Method for Speaker Tracking

在基于麦克风阵列的说话人跟踪系统中,存在观测方程的非线性程度较强,观测误差的统计特性不易准确描述等问题。本文提出了一种基于sigma点H_∞滤波的说话人跟踪方法。该方法用sigma点转换技术减小观测方程的线性化误差,用H_∞滤波方法降低观测误差不确定性对说话人位置估计的影响,从而提高了说话人跟踪精度,增强了跟踪系统对噪声的鲁棒性。仿真实验结果表明,相对于扩展卡尔曼滤波方法,本文方法在多种噪声条件下可将说话人跟踪误差降低25%以上。

Based on theories of H∞ filtering and sigma point transformation,a sigma point H∞ filtering method is proposed. Since sigma point transformation technique is used instead of Taylor series expansion in the new method, linearization error of high nonlinear system is decreased. Moreover,the noise uncertainty problem is solved by utilizing H∞ filtering method. Compared with existing method for microphone array based speaker tracking system, the estimation accuracy is improved and the system is more robust to measurement error of time difference of arrival （TDOA）. Simulation results of a speaker tracking system demonstrate that the proposed method decreases the tracking error by 25% compares to the extended Kalman filter.