摘要
针对强混响及噪声相关背景下说话人跟踪问题,提出了一种解相关粒子滤波跟踪方法。利用状态方程恒等变换和矩阵相似变换理论解除过程噪声和量测噪声以及测量噪声之间的相关性,在粒子滤波的框架内采用布朗模型对说话人运动进行建模,并将随机有限集(Random Finite Set,RFS)理论引入到说话人状态的归一化处理中,采用RFS将说话人的位置和说话人的数目综合成单个的变量集合,并在PF的框架内进行说话人跟踪实验。计算机仿真和实际的语料库跟踪实验验证了所提算法的有效性。
To deal with the speaker tracking problem in the background of strong reverberation and correlative noise,a de-correlation particle filter tracking method is proposed.The algorithm firstly uses the theory of state equation identical transformation and matrix similarity transformation to remove the process noise and measurement noise as well as the correlation between the measurement noise.And then within the framework of particle filter,Brown model is used to modeling the speaker motion and the Stochastic Finite Set(Random Finite Set,RFS)theory is introduced into the normalized processing of the speaker state.The RFS is used to integrate the location of the speaker and the number of speakers into a single set of variables,and the speaker tracking experiments are made within the framework of PF.Computer simulation and practical corpus tracking experiment verify the effectiveness of the proposed algorithm.
作者
杨海红
王琳娟
YANG Haihong;WANG Linjuan(Department of Computer Science,Shanxi Vocational College of Tourism,Taiyuan 030031,China;Department of Basic Courses,Shanxi Agricultural University,Jinzhong 030801,China)
出处
《无线电工程》
北大核心
2021年第9期963-970,共8页
Radio Engineering
基金
国家自然科学基金资助项目(61873153)。