English title

We present the new LICAD algorithm to solve the permutation problem of the ICA in the frequency domain and improve the separation quality. In the proposed algorithm, first, the sources' angles are estimated in each frequency bin using an ICA separating matrix. Then, these estimates are compared to the true values obtained from a pre-processing stage. If the difference among similar angles is less than a specified threshold, its corresponding permutation with the least difference is selected. Otherwise, the correlations among permutations of all ICA and LQBP outputs are calculated and the best permutation with the maximum correlation is chosen. By comparing the SDRs and SIRs, we demonstrate that the LICAD algorithm improves the separation quality of the conventional ICA