For Full-Text PDF, please login, if you are a member of IEICE,|
or go to Pay Per View on menu list, if you are a nonmember of IEICE.
RLS-Based On-Line Sparse Nonnegative Matrix Factorization Method for Acoustic Signal Processing Systems
IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences
Publication Date: 2013/05/01
Online ISSN: 1745-1337
Print ISSN: 0916-8508
Type of Manuscript: LETTER
Category: Engineering Acoustics
NMF, ALS-NMF, RLS, on-line NMF, sparse NMF,
Full Text: PDF>>
Recursive least squares-based online nonnegative matrix factorization (RLS-ONMF), an NMF algorithm based on the RLS method, was developed to solve the NMF problem online. However, this method suffers from a partial-data problem. In this study, the partial-data problem is resolved by developing an improved online NMF algorithm using RLS and a sparsity constraint. The proposed method, RLS-based online sparse NMF (RLS-OSNMF), consists of two steps; an estimation step that optimizes the Euclidean NMF cost function, and a shaping step that satisfies the sparsity constraint. The proposed algorithm was evaluated with recorded speech and music data and with the RWC music database. The results show that the proposed algorithm performs better than conventional RLS-ONMF, especially during the adaptation process.