Subband autoregressive modelling for speech enhancement

Authors

  • Brady Laska Dept. of Systems and Computer Engineering, Carleton University, Ottawa, Canada
  • Rafik Goubran Dept. of Systems and Computer Engineering, Carleton University, Ottawa, Canada
  • Miodrag Bolic University of Ottawa, School of Information Technology and Engineering, Ottawa, Canada

Keywords:

Argon, Control theory, Speech communication, Speech enhancement, AR models, Autoregressive modeling, Autoregressive modelling, High quality, High-order, Parametric forms, Residual noise, Speech enhancement algorithm, Speech signal enhancement, Speech signals, Sub-bands

Abstract

The use of subband autoregressive (AR) modeling for speech enhancement is discussed. The parametric form of the AR model provides an efficient and low-variance representation of the speech signal spectrum. This allows for significant compression benefits in speech communications and can be applied to speech signal enhancement. Speech enhancement algorithms make an effort to remove the additive noise without distorting the desired speech signal. Kalman filter speech enhancement provides high quality enhanced speech with natural sounding residual noise by enforcing an AR model structure. An alternative to using a single high-order AR model is to use a filterbank to decompose the speech and to model each subband channel with a significantly low-order AR model.

Published

2009-09-01

How to Cite

1.
Laska B, Goubran R, Bolic M. Subband autoregressive modelling for speech enhancement. Canadian Acoustics [Internet]. 2009Sep.1 [cited 2021Apr.13];37(3):62-3. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/2134

Issue

Section

Proceedings of the Acoustics Week in Canada