基于差分相位相关收敛准则的迭代语音信号分离
Iterative speech separation controlled by difference phase-correlation-based criterion function
-
摘要: 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。Abstract: A novel and improved iterative speech enhancement approach is presented to suppress the undesired speech or noise in cocktail party environments. The key idea in the proposed approach is that a difference phase- correlation-based criterion function(DPCF) is introduced to work as the convergence controller for a modified iterative wiener filter(MIWF). Moreover, an adaptive-filter-based method is employed to adaptively set the MIWF's parameters according the variance of environment. The interference component is attenuated continually until convergence occurs. Evaluation results of objective quality measurement and informal listening test show that the proposed system provides speech with good quality and good intelligibility.