Imperial College London > Talks@ee.imperial > COMMSP Seminar > Recent Advances in Bayesian STSA Estimation for Single-Channel Speech Enhancement

Recent Advances in Bayesian STSA Estimation for Single-Channel Speech Enhancement

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact d.mandic.

Single channel speech enhancement algorithms are commonly used to remove background noise in modern speech processing systems. They are present in many common devices such as cell phones, audio interface for automatic speech recognition and hearing aids. Much ground has been covered since the seminal work by Boll on spectral subtraction in the late seventies and several new speech enhancement approaches have been proposed and investigated, ranging from Wiener filter based approaches to the more sophisticated subspace methods. According to recent experimental studies, speech enhancement methods based on the estimation of the speech short-term spectral amplitudes (STSA) appear to offer the best compromise between residual (musical) noise and signal distortion. In the first part of the talk, we will give a detailed overview of single channel STSA methods of speech enhancement and summarize recent developments by other researchers, including the LSA , COSH, β-SA, WE and WCOSH approaches. SNR estimation and other modeling issues, including a recent trend toward the use of so-called super-Gaussian distributions, will also be discussed. In the second part of the talk, we will present some our most recent research work on single-channel Bayesian STSA estimation, including: 1) an improved weighted β-SA estimator with auditory based parameters that combine the power law of the β-SA and the weighting factor of the WE methods; 2) a simplified mathematical framework that allows the unification and generalization of several Bayesian STSA estimators; and 3) multi-dimensional extensions of STSA estimators that exploit the correlation between spectral components in frame-based speech processing to improve the enhancement performance. Evaluation results from subjective listening tests and objective (standardized) speech quality measures will be presented to support these findings.

Bio: Benoit Champagne was born in Joliette (P.Q.) Canada. He received the B.Ing. degree in Engineering Physics from the Ecole Polytechnique de Montréal in 1983, the M.Sc. degree in Mathematical Physics from the Université de Montréal in 1985, and the Ph.D. degree in Electrical Engineering from the University of Toronto in 1990. From 1990 to 1999, he was an Assistant, and then Associate Professor at INRS -Telecommunications, Université du Quebec, Montréal. In 1999, he joined McGill University, Montreal, as an Associate Professor within the Department of Electrical and Computer Engineering. He served as Associate Chairman of Graduate Studies in the Department from 2004 to 2007. His research focuses on the development and investigation of new computational algorithms for the processing of information bearing signals by digital means. His interests span many areas of statistical signal processing, including detection and parameter estimation, sensor array processing, adaptive filtering, and applications thereof to broadband voice/data communication systems. He has published several papers in these areas, including key works on subspace tracking, spread sources localization and time delay estimation. His research has been funded by the Natural Sciences and Engineering Research Council (NSERC) of Canada, the Fonds de recherche sur la nature et les technologies from the Government of Quebec, as well as some major industrial sponsors, including Nortel Networks, Bell Canada and InterDigital. He has been an Associate Editor for the IEEE Signal Processing Letters and the EURASIP Journal on Applied Signal Processing and has served on the Technical Committees of several international conferences. His is currently a Senior Member of IEEE

This talk is part of the COMMSP Seminar series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

Changes to Talks@imperial | Privacy and Publicity