A new mixing matrix estimation method for under-determined blind source separation of audio signals is proposed. By statistically modeling the local features, i.e., the magnitude ratio and phase difference of the mixtures, in a time-frequency region, a region can have information of the mixing angle of a source with reliability amounted to its likelihood. Regional data are then clustered with statistical tests based on their likelihood to produce estimates for the mixing angle of the sources as well as the number of them. Experimental results show that the proposed mixing matrix estimation algorithm outperform the existing methods.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.