论文部分内容阅读
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。
A speech separation algorithm based on talker state detection is proposed. The algorithm detects the state of the interlocutor automatically and controls the adaptive filtering process according to the corresponding state, so as to estimate the sound field transfer function of each path, and then the mixed speech signal is separated. The simulation results show that the proposed algorithm overcomes the defect that the reference signal impure leads to the deterioration of adaptive speech separation compared with the traditional signal separation algorithm. This algorithm does not need to artificially reduce the adaptive filter So it has faster convergence and tracking performance. In addition, the algorithm has the advantages of less computation and good real-time performance.