基于独立分量分析与小波包分解的混叠声源信号波形恢复

Waveform Restoral of Multiple Mixed Acoustical Source Signals Based on Independent Component Analysis and Wavelet Packet Decomposition

通过联合应用独立分量分析与小波包分解,提出一种混叠声源信号波形恢复的新方法。首先,借助基于奇异值分解的聚类分析估计原始多通道混合声观测中的独立源数。在此基础上,依据最小相关准则削减原始观测维数。随后,利用独立分量分析的冗余取消特性,由被削减的新观测出发初步抽取各独立源分量。最后,基于小波包分解进行信号消噪,实现多个混叠声源信号的波形恢复。实验结果表明,基于独立分量分析与小波包分解联合的新方法,能有效分离并正确恢复被噪声强烈污染的混叠声源信号波形,从而为后续的应用(如声源识别、声学故障诊断等)奠定了基础。

A new method for acoustical sources waveform restoral was proposed,by the joint use of ICA and WPD.First,the number of incoherent sources in a mixing system was estimated by clustering analysis based on singular value decomposition.Furthermore,original multi-channel acoustical observations were dimensionally reduced according to minimum correlation criterion.Then,independent source components were primarily extracted from the reduced observations,by the ICA based redundancy reduction.Finally,WPD technique was applied to denoise the extracted source signals.Thus,multiple mixed acoustical sources were separated and their waveforms were restored.Experimental results indicate that the new ICA-WPD based method enable effective separation and correct waveform restoral of acoustical source signals which are mixed and corrupted by strong additive noise,and establishment of a solid base for some succedent applications such as acoustical sources recognition and acoustic diagnostic,etc.