Within-Class Covariance Correction (WCC) for Linear Discriminant Analysis (LDA) in Speaker Recognition
, BBN Technologies
Date: Monday, April 28, 2014
Time: 3:00 PM to 4:00 PM Note: all times are in the Eastern Time Zone
Refreshments: 2:45 PM
Location: 32-G882 (Stata Center - Hewlett Room)
Host: Jim Glass and Najim Dehak, MIT CSAIL
Contact: Marcia G. Davidson, 617-253-3049, email@example.com
Speaker URL: None
TALK: Within-Class Covariance Correction (WCC) for Linear Discriminant Analysis (LDA) in Speaker Recognition
This talk presents the technique of Within-Class Covariance Correction (WCC) for Linear Discriminant Analysis (LDA) in Speaker Recognition to perform an unsupervised adaptation of LDA to an unseen data domain, and/or to compensate for speaker population difference among different portions of LDA training dataset. The work follows on the study of source-normalization and interdatabase variability compensation techniques which deal with multimodal distribution of i-vectors. On the Domain Adaptation Challenge (JHU 2012 Workshop) we show up to 70% relative EER reduction and we propose a data clustering procedure to identify the directions of the domain-based variability in the adaptation data. On the DARPA RATS (Robust Automatic Transcription of Speech) task, we show that, with two hours of unsupervised data, we improve the Equal-Error Rate (EER) by 17.5%, and 36% relative on the unmatched and semi-matched conditions, respectively.
Ondrej Glembek received his Ph.D. degree (2011) in signal, image, and speech processing from the Department of Computer Graphics and Multimedia, Faculty of Information Technology, Brno University of Technology. The topic of his Ph.D. was "OPTIMIZATION OF GAUSSIAN MIXTURE SUBSPACE MODELS AND RELATED SCORING ALGORITHMS IN SPEAKER VERIFICATION" and his main interest has ever since then been Robust Speaker and Language Recognition. Ondrej took part in all NIST Language and Speaker evaluation since 2006 as a member of the BUT team. He has also participated in various international projects such as EU MOBIO, and IARPA BEST. He has also participated in JHU 2008 Summer workshop on Robust Speaker Recognition, as well in the following BOSARIS Speaker Recognition workshop in later years. In 3/2013, he started a position at Raytheon BBN Technologies as a post-doc researcher, working on the Robust Automatic Speech Transcription (RATS) DARPA program. At 2011 ICASSP conference, Ondrej received the Ganesh Ramaswamy (IBM) prize for his paper "Simplification and Optimization of i-vector extraction."
Created by Marcia G. Davidson at Wednesday, April 23, 2014 at 12:33 PM.