A Signal Separation Method Based on Sparsity Estimation of Source Signals and Non-negative Matrix Factorization
In order to improve the signal separation performance of the non-negative matrix factorization (NMF), sparse
non-negative matrix factorization (Sparse NMF, SNMF) was developed. Existing SNMF algorithm uses arbitrarily
determined sparseness without considering the sparseness of individual sound sources. In this paper, we propose a new
signal separation method that estimates the sparseness according to the characteristics of a sound source and applies it to
SNMF algorithm. Experimental results show that the proposed method has better performance than the existing NMF and
Keywords - Signal separation, Non-negative matrix factorization, Sparseness, Denoising.