On-line nonnegative matrix factorization for music signal separation

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

Nonnegative matrix factorization (NMF) is widely used in audio source analysis and audio signal separation systems. However, conventional NMF algorithms cannot be used on-line because they process all available data at once. On-line NMF algorithms have been developed recently, but most of them cannot be applied to audio systems. In this paper, an on-line NMF algorithm, based on multiple normal equations and recursive solutions, is developed, and the developed algorithm is extended to more enhanced algorithm and more simplified algorithm. An on-line monaural ambient signal separation system is presented to demonstrate the applicability of the proposed algorithm.

Original languageEnglish
Title of host publicationProceedings - 2014 10th International Conference on Intelligent Information Hiding and Multimedia Signal Processing, IIH-MSP 2014
EditorsJunzo Watada, Akinori Ito, Jeng-Shyang Pan, Han-Chieh Chao, Chien-Ming Chen
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages566-567
Number of pages2
ISBN (Electronic)9781479953905
DOIs
StatePublished - 24 Dec 2014
Event10th International Conference on Intelligent Information Hiding and Multimedia Signal Processing, IIH-MSP 2014 - Kitakyushu, Japan
Duration: 27 Aug 201429 Aug 2014

Publication series

NameProceedings - 2014 10th International Conference on Intelligent Information Hiding and Multimedia Signal Processing, IIH-MSP 2014

Conference

Conference10th International Conference on Intelligent Information Hiding and Multimedia Signal Processing, IIH-MSP 2014
Country/TerritoryJapan
CityKitakyushu
Period27/08/1429/08/14

Keywords

  • audio source separation
  • nonnegative matrix factorization
  • on-line audio signal processing
  • recursive least squares

Fingerprint

Dive into the research topics of 'On-line nonnegative matrix factorization for music signal separation'. Together they form a unique fingerprint.

Cite this