摘要
针对强噪声背景下的说话人跟踪系统,提出了一种适应于噪声统计特性未知的无迹扩展H∞粒子滤波方法,并将其应用于强噪声背景下的说话人跟踪问题.首先,将无迹变换(UT)变换引入到适用于噪声统计特性未知的EHF中取代复杂的雅克比矩阵计算,降低观测方程线性化引起的误差;接着,采用生成的无迹扩展H∞滤波优化重要性概率密度函数,将最新观测信息引入到粒子修正过程;最后,对本方法的粒子采样实现和权重更新步骤进行了详细分析.仿真分析表明:本方法有效提升了在信噪比较低情况下的跟踪精度和鲁棒性.
This paper proposed a new kind of extended H∞ infinity particle filter algorithm which adapts to the situation of unknown noise statistical characteristics.The method was used to the speaker tracking problem under the strong noise background.Firstly,unscented transform(UT)was introduced to extended H∞infinity filter to replace complex Jacobi matrix,which reduced the observation equations linearization error.Secondly,the proposed method was used to optimize the importance probability density function of particle filter,which introduced the latest observation information into particle modification process.Finally,the particle sampling and weights updating step was analyzed in detail.The simulation results show that the proposed method has improved speaker tracking system accuracy and robustness in the condition of low signal noise ratio(SNR).
出处
《华中科技大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2015年第S1期363-366,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(61263031)
甘肃省自然科学基金资助项目(1010RJZA046)
关键词
说话人跟踪
粒子滤波
H∞滤波
噪声统计特性
无迹变换
speaker tracking
particle filter
H∞infinity filter
noise statistical properties
unscented transform