考虑粒子记忆传播的说话人精确跟踪方法  

Accurate speaker tracking method considering particle memory propagation

作　　者：李卓[1] 李欣[2] Li Zhuo;Li Xin(Chengdu Normal University,Chengdu 610066,China;Sichuan Normal University,Chengdu 610066,China)

机构地区：[1]成都师范学院,成都611130 [2]四川师范大学,成都610066

出　　处：《国外电子测量技术》2022年第6期31-37,共7页Foreign Electronic Measurement Technology

基　　金：四川省科学技术厅2022年重点研发计划项目(重大科技专项)(22ZDYF2726);四川省高等学校人文社会科学重点研究基地·四川中小学安全教育与管理研究中心(ZXXAQ21-15)项目资助。

摘　　要：针对配备了摄像机和麦克风阵列的智能环境中说话人跟踪问题,提出了一种考虑运动说话人状态记忆效应的跟踪方法。该方法在粒子滤波框架内,融合说话人动态模型,通过粒子状态在时间上的连续预测克服虚假声源对跟踪精度的影响;通过借鉴粒子群算法中粒子的交互能力,通过优化适应度函数,引入适应度衰减机制,实现粒子记忆状态的跨时间传递,确保在发生说话人动态模型不匹配或当前观测错误的情况下,实现过往最佳匹配信息的记忆传递,并通过粒子相互作用,共享适应度信息,加快收敛速度,有效改善了粒子的收敛性能。实验结果表明,方法较好改善了不同复杂环境下的跟踪性能。Aiming at the problem of speaker tracking in intelligent environment equipped with camera and microphone array,a tracking method considering the state memory effect of moving speaker is proposed.In the framework of particle filter,the speaker dynamic model is fused,and the influence of false sound source on tracking accuracy is overcome through the continuous prediction of particle state in time.By referring to the interaction ability of particles in particle swarm optimization algorithm,optimizing the fitness function and introducing the fitness attenuation mechanism,the cross time transmission of particle memory state is realized,so as to ensure the memory transmission of the best matching information in the past in case of speaker dynamic model mismatch or current observation error,and share the fitness information through particle interaction,The convergence speed is accelerated and the convergence performance of particles is effectively improved.Experimental results show that the proposed method improves the tracking performance in different complex environments.

关 键 词：人工智能 说话人跟踪 粒子滤波 麦克风阵列 粒子群优化算法 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TH691.9[自动化与计算机技术—计算机科学与技术]