一种基于部分信道估计的DTD算法

A new double-talk detection algorithm based on partial channel estimate

在声学回声消除(acoustic echo cancellation,AEC)系统中,双讲(doubletalk,DT)信号和信道突变都导致滤波器误差变大,然而AEC系统需要分别对这2种情况做出不同的反映。因此,常见的基于误差信号及其函数的双讲检测(doubletalk detection,DTD)算法在信道突变情况下容易产生误判。为此,利用声学回声信道的特殊性和归一化最小均方(normalized least mean square,NLMS)算法的跟踪特性,提出了一种新型的基于部分信道估计的DTD算法。该算法采用累加信道估计尾部的值作为判决条件,能够区分出通话过程中的双讲时段并且忽略回声信道突变的影响。该算法具有计算复杂度小和检测DT信号速度快的优点。仿真结果表明,在实验条件下该算法优于常见的DTD算法。

In acoustic echo cancellation （AEC） system, both double-talk （DT） signal and acoustic channel change will in- crease estimation error. The AEC system needs to respond to these two cases differently. Therefore, common double-talk detection （DTD） algorithms based on error signal or its functions are prone to misjudgments when the acoustic channel sud- denly changes. By employing the special characteristics of acoustic echo channel and tracking properties of the normalized least mean square （NLMS） algorithm, a new DTD algorithm based on partial channel modeling is proposed. Using a deci- sion factor calculated from accumulating the values of the rear part of channel estimate, the proposed algorithm is able to distinguish DT period from acoustic echo channel change. It has low computational complexity and fast DT detection speed. Finally, simulations show that under experimental conditions the performance of proposed algorithm is better than those common algorithms.