采用聚类神经网络与分离输出语音重构的语音分离算法

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

Speech Separation Algorithm Using Clustering Neural Network and Speech Reconstruction

Based on the principle of computational auditory scene analysis (CASA), a CASA speech separation algorithm was proposed by using oscillator neural network (ONN). One example was used to explain the processing and the coefficients setting in the proposed algorithm. The second part of the proposed algorithm was discussed, which includes the clustering oscillator neural network and the speech separation outputs' reconstruction processing. The designation of the clustering ONN was discussed in detail by using appropriate auditory perceptive segments' clustering principles. The simulation results demonstrate that the proposed algorithm can achieve the speech separation successfully.