This paper proposes two geometric source separation methods. The first one is composed of a NMF-based separation module and clustering module with spatial information. The purpose of the algorithm is twofold: first, to enhance the interference injection performance relative to conventional beamformer algorithms, and second, to provide a useful clustering algorithm for decomposed bases after using the NMF method. The second proposed algorithm is composed of a beamformer module, a NMF-based separation and clustering module, and a selection module. The proposed system is compiled to assure robustness with the difference of the incidence angles between the desired signal and the interference signal. The evaluation was performed with recorded speech signals and the results are illustrated.
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.