A geometric approach for separating several speech signals 

Massoud Babaie-Zadeh 1,2, Ali Mansour 3, Christian Jutten 4, and Farrokh Marvasti 1,2 .


1 Multimedia Lab,Iran Telecom Research Center(ITRC),Tehran,Iran 

2 Electrical Engineering Department, Sharif University of Technology, Tehran, Iran 

3 E3I2,ENSIETA,Brest,France 

4 Institut National Polytechnique de Grenoble(INPG),Laboratoire des Images 
et des Signaux(LIS), Grenoble, France 



Abstract. In this paper a new geometrical approach for separating 
speech signals is presented. This approach can be directly applied to 
separate more than two speech signals. It is based on clustering the 
observation points,and then fitting aline(hyper-plane) ontoeach cluster. 
The algorithm quality is shown to be improved by using DCT coefficients 
of speech signals, as opposed to using speech samples.