DocumentCode :
1699578
Title :
Subband blind source separation for convolutive mixture of speech signals based on dynamic modeling
Author :
Mosayebi, Reza ; Sheikhzadeh, H. ; Raahemifar, Kaamran
Author_Institution :
Dept. of Electr. Eng., Amirkabir Univ. of Technol., Tehran, Iran
fYear :
2013
Abstract :
In this paper, a subband blind source separation method based on dynamic modeling for convolutive mixture of speech signals is proposed. We show that by applying a dynamical model to subband signals, some of the drawbacks of the time domain approach can be resolved, leading to improvements in separation performance. By employing the subband processing, we enhance the speed of the method, first by reducing the computational cost of the algorithm resulting from shorter demixing filters and second, by considering the parallel processing capability of the subband domain. Furthermore, by applying particular settings to the step-size parameter and to the demixing filter lengths in various subbands, we achieve much better performance in terms of the separation ability. The proposed algorithm is applied to two different experiments and a comparison is done against the time domain approach. The results demonstrate the superiority of the subband domain in terms of speed and accuracy.
Keywords :
blind source separation; filtering theory; speech processing; time-domain analysis; convolutive mixture; demixing filter lengths; dynamic modeling; parallel processing capability; separation ability; speech signals; subband blind source separation; subband processing; subband signals; time domain approach; Microphones; Convolutive blind source separation; demixing filter; dynamic modeling; subband domain;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Signal Processing and Information Technology(ISSPIT), 2013 IEEE International Symposium on
Conference_Location :
Athens
Type :
conf
DOI :
10.1109/ISSPIT.2013.6781897
Filename :
6781897
Link To Document :
بازگشت