摘要
提出了一种基于归一化互相关原理的双端通话检测器。它与目前广泛应用的基于线性预测原理的声码器相结合,直接从解码过程中提取短时激励信号作为去相关的参考信号以及线性预测系数用于对误差信号进行去相关。利用去相关后的参考信号和误差信号来构建归一化互相关检测变量,从而使计算复杂度从O(N2)降低至O(N)。仿真结果表明,该算法在双端通话和回声路径改变时检测准确,与去相关前的算法相比,在误报概率和漏报概率方面也有明显改善。
A novel doubletalk detector based on normalized cross-correlation principle is developed. By combining with the widely used LPC-based speech coders, it extracts the short-term excitement signal used as the whitened reference signal and the linear prediction coefficients are used to whiten the error signal directly from the decoding process. Two whitened signals are used to compute the normalized cross-correlation detection variable, which reduces the compu- tational complexity from O（N^2） to O（N）. Simulation results demonstrate that the algorithm detects correctly the change in both doubletalk and echo-path. Meanwhile, the probabilities of both the false alarm and the miss alarm are improved compared with the algorithms before decorrelation.
出处
《数据采集与处理》
CSCD
北大核心
2007年第3期257-261,共5页
Journal of Data Acquisition and Processing
基金
国家自然科学基金(60572042)资助项目
关键词
回声消除
双端通话
互相关
去相关
echo cancellation
doubletalk
cross-correlation
de-correlation
