A Generalized Stein's Estimation Approach to Speech Enhancement Based on Perceptual Criteria

Sunder Ram Krishnan

Indian Institute of Science, Bangalore

Tuesday, September 11, 2012
12:30 p.m.

We develop estimators for risks corresponding to non-quadratic, perceptually motivated Itakura-Saito (IS) distortion measure within the Stein's formalism. The observation model assumed is general, in the sense that the noise pdf is considered to belong to the exponential family of densities, with there being the possibility of correlation as well. The work assumes significance considering the facts that enhancement can now be performed in a general setting, for various models, and that previous works have only focussed on Stein's risk estimator development for MSE in a general model and IS risk in an i.i.d. Gaussian setting. We show some enhancement results with correlated Gaussian noise.

Bio:
Sunder Ram Krishnan received his B.Tech. degree in Electronics and Communication Engineering from the College of Engineering, Thiruvananthapuram, Kerala, India, which he passed in First Class with Distinction. He is currently pursuing the M.S. degree in Signal Processing at the Indian Institute of Science (IISc), Bangalore, under the guidance of Dr. Chandra Sekhar Seelamantula at the Department of Electrical Engineering. His research focusses on developing risk estimation-based frameworks for non-parametric regression problems. A paper titled "A Generalized Stein's Estimation Approach to Speech Enhancement Based on Perceptual Criteria" has been accepted to SAPA-SCALE 2012, and two of his papers have already been accepted to the Journal of Optical Society of America ("A zero-Crossing Approach to High-Resolution Reconstruction in Frequency-Domain Optical-Coherence Tomography") and to the IEEE Transactions on Signal Processing ("On the Selection of Optimum Savtzky-Golay Filters").