基于话者状态检测的自适应语音分离方法的研究

提出了一种基于话者状态检测的语音分离算法。该算法对话者状态进行自动检测,并根据相应的状态对自适应滤波过程加以控制,以此对各路的声场传递函数进行估计,进而使混合的语音信号得到分离。仿真实验结果表明:与传统的输出信号互为参考的信号的分离算法相比,该算法克服了参考信号不纯导致自适应语音分离结果恶化的缺陷;该算法不需要人为地降低自适应滤波器的收敛速度,所以具有较快的收敛和跟踪性能;此外,该算法还具有运算量较小,实时性好等特点。

Research of adaptive speech separation method based on speech status detection

A new adaptive signal processing scheme based on the detection of the speech status is proposed to solve the problem of speech separation. The proposed method automatically determines the speech status of different speakers efficiently and controls the adaptive processing according to the different speech status, therefore, the transfer function of each acoustic path can be estimated more precisely, leading to improved speech separation results. The simulation results demonstrate the superiority of the proposed algorithm over the conventional method which uses the mixed signals as the reference signal mutually, and the main reason is that the deterioration of the speech separation results caused by the impurity of the reference signal is avoided. Furthermore, the real-time implementation of the algorithm is promising because there is no necessity of reducing the convergence rate to guarantee the stability and the computation burden increase is within a tolerable range.