Audio source indexing by sub-band PCA with empirical mode decomposition

This paper presents a sub-bund approach of audio source indexing technique without any prior information about the sources. The empirical mode decomposition (EMD) scheme, capable of decomposing nonlinear and non-stationary signal into some bases, is employed to implement the proposed sub-band method. The feature vectors are derived from each of the selected sub-bands of the signal block. Linear predictive cepstral coefficient (LPCC) is used as the principal feature of this indexing system. The dominant features are selected at each sub-band with principal component analysis (PCA). A dominancy dependent weighted function is introduced to measure the similarity between the source model and the derived feature vectors. The higher order statistics (HOS) is employed to compute the sub-band LPCC features. The use of HOS makes LPCC less affected by Gaussian type noises. The experimental results show that the sub-band technique produces better indexing efficiency than that of the full-band technique.