Marina-Anca CIDOTA
Blind Separation for Speech Signals by Minimizing the Contingency Coefficient in the Chi-Square Independence Test

Abstract. In this paper, the problem of blind source separation (BSS) using an independence measure for observations is addressed. Random variables are transformed by their respective estimate distribution functions into uniform random variables, whose independence is evaluated using the contingency coefficient from the chi-square independence test. A new objective function for the adaptive blind source separation algorithm is thus proposed. Simulations illustrate that this algorithm based on minimizing the contingency coefficient leads to reliable results when applied to separate speech signals which are linearly mixed.