In spatial audio analysis-synthesis, one of the key issues is to decompose a signal into primary and ambient components based on their spatial features. Principal component analysis (PCA) has been widely employed in primary component extraction, and shifted PCA (SPCA) is employed to enhance the primary extraction for input signals involving the inter-channel time difference. However, SPCA generally requires the primary components to come from one direction and cannot produce good results in the case of multiple directions. To solve this problem, we propose multi-shift PCA (MSPCA) by extending SPCA to multiple shifts. Two structures of MSPCA with different weighting methods are discussed. From the results of our simulations and listening tests, the proposed consecutive MSPCA with proper weighting is found to be superior to the conventional PCA and SPCA based primary extraction methods.

This paper will be presented in ICASSP 2015.