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。
This paper shows the importance of the optimal smoothing scheme in Microphone Array Post-Filtering(MAPF) under a combined Deterministic-Stochastic Hybrid Model(DSHM).We reveal that some of the well-known MAPF algorithms may cause serious speech distortion without using the optimal smoothing scheme,which is resulted from oversmoothing the raw periodogram over time.Using a minimum conditional mean square error criterion,we derive the optimal smoothing factor under the DSHM,where the Deterministic-to-Stochastic-Ratio(DSR) and the stationarity determine the value of the optimal smoothing factor.The optimal smoothing scheme is applied to the Tran-sient-Beam-to-Reference-Ratio(TBRR)-based MAPF algorithm and experimental results show its better performance in terms of both the Log-Spectral Distance(LSD) and the Perceptual Evaluation of Speech Quality(PESQ).
