一种基于半盲源分离的回声抵消方法

An Echo Cancellation Method Based on Semi-blind Source Separation

传统的声学回声抵消算法中双端对讲会导致自适应滤波器的更新受到影响,因此一般需要配合双端对讲检测技术来调节滤波器的更新速度。本文提出了一种基于半盲源分离的无需双端对讲检测的频域回声抵消方法。假设近端和远端信号为零均值复高斯分布的前提下,在半盲源分离问题的框架中采用最大似然估计得到回声路径,在估计回声路径的更新过程中引入对近端信号的能量估计,解决了近端信号对回声路径估计的干扰问题。实验表明,该方法在双端对讲、路径变化以及不同信噪比下均表现出较好性能。

The performance of traditional acoustic echo cancellation technique usually suffers from the near-end signal interference during double-talk conditions.To deal with this problem,double-talk detection(DTD)techniques are used to adjust the rate of convergence.An algorithm based on semi-blind source separation is proposed for acoustic echo cancellation in the short-time Fourier transform domain to learn echo paths without double-talk detector.In the framework of semi-blind source separation,the maximum likelihood estimation is used to learn the echo path under the assumption that the near-end and far-end signals follow a zero-mean Gaussian distribution.The energy estimation of near-end signals is introduced to the process of learning echo paths,and hence the near-end interference problem is resolved.Experiments show that the proposed method has good performance in the presence of double-talk,path changes or different signal-to-noise ratio.