摘要
双端发声检测是回声抵消应用中的一项关键技术,如果检测错误,会影响自适应滤波算法的收敛速度,严重时甚至会引起发散,因而,如何构造精确的双端检测算法已成为整个技术发展的瓶颈。目前,常用的几种检测方法各自存在优缺点。通过搭建仿真链路,模拟双端放声状况,对这些常用算法的性能进行了分析比较。由分析可得出,虽然双滤波器法的性能最优,但是存在计算量偏大的问题;提出了一种新的算法—系数缓存法,它仍然具——有回复收敛的特质,但计算量降低近一半。
The double-talk detection is a key part in the echo-cancellation. If the detection fails, adaptive-filter's performance will be reduced largely. So, how to construct a DTD algorithm has become the battle-neck. Now, there are various problems in several DTD algorithms. Through building simulation linkage to simulate DTD case, the performances of these algorithms are analyzed and compared. Based on these comparisons, the double-filter has the best performance, but its calculation is too large. A new algorithm coefficient-memory which can return convergence is proposed. The calculation of this new algorithm is reduced nearly to 50%.
出处
《电声技术》
2008年第6期68-71,76,共5页
Audio Engineering
基金
国家"863"计划引导项目(2004AA001390)
国家高技术研发发展计划(863)课题(2004AA123150)