A Frequency-domain Blind Source Separation and Ranking Algorithm Based on the Influence of Frequency Points
A frequency point-influenced convolutional hybrid blind source separation and rank-ing algorithm is proposed to address the influence of ranking ambiguity on convolutional hybrid blind source separation performance.In the improved algorithm,the separation matrix of fre-quency points is measured using weight coefficients,and then used as the iterative initial value of the next separation matrix to maintain the continuity of the determinant of the separation ma-trix.The distance of frequency points is controlled by an influence factor to improve the quality of frequency point separation and rank the frequency points based on reliable frequency points,so as to improve the estimation accuracy of the separation algorithm.The similarity coefficient,SIR,SAR and SDR performance indexes of the algorithm are improved in the simulation experi-ments for blind source separation of indoor speech signals,which proves the effectiveness of fre-quency point influence.
blind source separationconvolutive mixturefrequency domain rankingfrequency point influence
万方数据
维普
