Blind separation of convolutive speech mixtures with background interference employing a Hybrid approach With ICA & PCA.

Abstract

This paper presents a comparative analysis of Principal component analysis (PCA) and a rapidly emerging novel method, that is, the independent component analysis (ICA).An algorithm is designed in order to segregate the convolutive mixtures of speech with background noise utilizing two microphones recordings. The efficiency of the target speech has been analyzed by utilizing principle component analysis (PCA) and ideal binary mask (IBM), proceeding by post-filtering in cepstral domain. The segregation of the speech signal with background noise is achieved by three steps algorithm. Initially PCA algorithm is applied on mixture of source signals received by two microphone recording to segregate them. In the next step, the segregated sources obtain by PCA is used to assess the IBM with the comparison of the energy of corresponding time-frequency (T-F) units. Finally, T-Fmasking using cepstral smoothing is used to decrease musical noise. Then the results achieved using PCA based approach has been compared with ICA based approach.