论文部分内容阅读
将一种新颖的迭代算法用于消除语音信号中非目标话者的语音干扰。一个简单的双传声器系统被用来采集混合语音，形成迭代维纳滤波器的主通道信号和参考信号。算法的关键在于提出差分相位相关准则（ＤＰＣＦ）控制造代过程的收敛，并根据话者状态多类判决实时更新滤波器以自动适应声学环境．实验表明，迭代算法不断的消除主通道中的干扰话音直到准则函数判断收敛．客观语音评价和非正式的试听评价表明系统在达到收敛的同时，目标语音也达到最佳的语音质量，清晰度令人满意。
A novel iterative algorithm is used to eliminate the speech interference of non-target speakers in speech signals. A simple dual microphone system is used to acquire mixed speech to form the main channel signal and reference signal for the iterative Wiener filter. The key point of the algorithm is to propose the Differential Phase Correlation Criterion (DPCF) to control the convergence of the generation process, and to update the filter according to the different states of the speaker state to automatically adapt to the acoustic environment. Experiments show that the iterative algorithm keeps eliminating the interfering speech in the main channel until the criterion function converges. Objective speech evaluation and informal audition evaluation show that the system achieves the best speech quality with satisfactory clarity while achieving convergence.