A Puzzle of No Return? (part two)

jigsaw_puzzle.jpgWhat I mean more generally is this: To what extent can I expect a specific sparse approximation process and dictionary to hinder the separation of several sources from a mixture decomposed into atoms? Will a sparse approximation of a mixture make impossible an agglomeration of atoms into the separated sources? More specifically, can I expect to take a matching pursuit decomposition of a speech and music mixture using an overcomplete scale-time-frequency dictionary, and reconstruct the two sources using disjoint subsets of the book? This is like taking a jigsaw puzzle of a nature scene and building up just the trees, or just the clouds, from the pieces.

It is a naive assumption that each atom belongs to either the speech or music signal, especially for greedy pursuits that are prone to represent errors made in the model rather than anything real in the signal. But I wonder how bad of an assumption it is. Looks like I have got a new reading list for Paper of the Day (Po'D) next week!

N. Cho, Y. Shiu, and C.-C. J. Kuo, "Audio source separation with matching pursuit and content-adaptive dictionaries (MP-CAD)," in Proc. IEEE Workshop Appl. Signal Process. Audio Acoust., (New Paltz, NY), pp. 287-290, Oct. 2007.

B. V. Gowreesunker and A. H. Tewfik, "Blind source separation using monochannel overcomplete dictionaries," in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process., (Las Vegas, NV), pp. 33-36, Mar. 2008.

R. Gribonval, "Sparse decomposition of stereo signals with matching pursuit and application to blind separation of more than two sources from a stereo mixture," in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process., vol. 3, (Orlando, FL), pp. 3057-3060, May 2002.

S. Lesage, S. Krstulovic, and R. Gribonval, "Underdetermined source separation: Comparison of two approaches based on sparse decompositions," in Proc. Int. Conf. Independent Component Analysis Blind Source Separation, (Charleston, South Carolina), pp. 633-640, Mar. 2006.

P. Sugden and N. Canagarajah, "Underdetermined noisy blind separation using dual matching pursuits," in Proc. IEEE Int. Conf. Acoustics, Speech, Signal Process., (Montreal, Quebec, Canada), pp. 557-560, May 2004.   
Reblog this post [with Zemanta]

CRISSP is a research group in ADMT at Aalborg University Copenhagen (AAU-KBH), Denmark.

Authors:

  Bob L. Sturm
  Sofia Dahl
  Stefania Serafin

Archives

CRISSP @ Medialogy

↑ Grab this Headline Animator

Powered by Movable Type 4.34-en