Abstract

Our ears often simultaneously receive various sound sources (speech, music, noise . . .), but we can still listen to the intended sound. A system of speech recognition must be able to achieve the same intelligent level. The problem is that we receive many mixed (combined) signals from many different source signals, and would like to recover them separately. This is the problem of Blind Source Separation (BSS). In the last decade or so a method has been developed to solve the above problem effectively, that is the Independent Component Analysis (ICA). There are many ICA algorithms for different applications. This report describes our application to sound separation when there are more sources than mixtures (underdetermined case). The results were quite good.