基于听觉现象分析计算模型 ( CASA)的基本原理 ,对仅有单通道输入混合语音信号时 ,采用振荡器神经网络 ,提出了一种 CASA计算模型语音分离算法结构 .利用实例说明了算法的具体实现步骤和参数设置 .讨论了该算法结构中各语音听觉感知成分 Segments的聚类过程和对分离输出语音的重构处理部分 ,以及如何采用合适的听觉感知成分聚类规则设计相应的聚类神经网络 ,以完成对应不同输入独立语音源信号的各 Segments的聚类 ,从而实现语音分离任务 Based on the basic principle of CASA, a structure of speech separation algorithm based on CASA is proposed by using oscillator neural network when only single-channel input speech signal is mixed. The concrete implementation of the algorithm is illustrated by an example Steps and parameters setting.We discuss the clustering process of Segments and the reconstructed processing of the segmental output speech in the structure of the algorithm and how to design the corresponding clustering neural network using the appropriate clustering rules of auditory perception components , So as to complete the clustering of Segments corresponding to different input independent speech source signals so as to realize the task of speech separation