摘要
利用分层采样方法,融合波达方向和时间延迟两种信息,实现了对说话人的定位与跟踪.分层采样方法考虑波达方向和时间延迟这两种不同观测信息对说话人位置估计精度的差异,将基于波达方向滤波得到的状态后验概率密度函数作为基于时间延迟滤波的重要性采样函数,增强了重要性概率密度函数与后验概率密度函数的相似程度,从而改善了重要性概率密度函数的质量,减小了采样粒子权值的方差,提高了对说话人位置的估计精度.仿真实验验证了该方法的有效性.
Utilizing layered sampling method, both direction of arrival （DOA） and time difference of arrival （TDOA） of speech source are fused to localize and track the speaker. Since the measurement modalities differ in the level of information which they provide about the state, the layered sampling method constructs an informed proposal by integrating filtering results from DOA measurement, then particles are sampled from this proposal in TDOA measurement based particle filter. As the similarity between the importance density function and the posterior density function is enhanced, the quality of the proposal function is improved and variance of sample weights is decreased, thus the speaker localization accuracy is improved. Simulation results of two scenarios show the validity of the proposed method.
出处
《大连理工大学学报》
EI
CAS
CSCD
北大核心
2009年第4期580-586,共7页
Journal of Dalian University of Technology
基金
国家自然科学基金资助项目(60772161
60372082)
高等学校博士学科点专项科研基金资助项目(200801410015)
作者简介
侯代文（1972-），男，博士，E-mail：hodevin@gmail．com
殷福亮（1962-），男，教授，博士生导师，E-mail：flyin@dlut．edu.cn.
