A recursive least-squares extension of the natural gradient algorithm for blind signal separation of audio mixtures

Authors

  • M. Elsabrouty Sch. of Info. Technol. and Eng., University of Ottawa, 800 King Edward, Ottawa, Ont. K1N 6N5
  • T. Aboulnasr Sch. of Info. Technol. and Eng., University of Ottawa, 800 King Edward, Ottawa, Ont. K1N 6N5
  • M. Bouchard Sch. of Info. Technol. and Eng., University of Ottawa, 800 King Edward, Ottawa, Ont. K1N 6N5

Keywords:

Approximation theory, Audio systems, Computer simulation, Convergence of numerical methods, Matrix algebra, Maximum likelihood estimation, Natural frequencies, Parameter estimation, Probability density function, Audio mixtures, Blind signal separation (BSS), Gradient algorithms, Recrusive-Least-Squares (RLS)

Abstract

An algorithm named the Quasi-RLS Stiefel, that combines the principles of natural-gradient on different manifolds and RLS-based algorithms, was presented for blind signal separation (BSS) of audio mixtures. A mixture of audio data files sampled at 8 kHz and of duration 3.7 seconds were tested to compare the different BSS algorithms. It was observed that the new algorithm converges much faster than the other algorithms. It was found by the simulation results that the algorithm was suitable for BSS.

Additional Files

Published

2004-09-01

How to Cite

1.
Elsabrouty M, Aboulnasr T, Bouchard M. A recursive least-squares extension of the natural gradient algorithm for blind signal separation of audio mixtures. Canadian Acoustics [Internet]. 2004 Sep. 1 [cited 2024 Dec. 8];32(3):136-7. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/1665

Issue

Section

Proceedings of the Acoustics Week in Canada