Vlasios Doumpiotis
Johns Hopkins University
8 Papers
81 Citations
Vlasios Doumpiotis is an academic researcher from Johns Hopkins University. The author has contributed to research in topics: Discriminative model & Hidden Markov model. The author has an hindex of 6, co-authored 8 publications. Previous affiliations of Vlasios Doumpiotis include Technical University of Crete.
Chat about Author
Papers
Discriminative linear transforms for feature normalization and speaker adaptation in HMM estimation
TL;DR: In this article, discriminative training procedures that employ linear transforms for feature normalization and for speaker adaptive training were introduced for improving large vocabulary conversational speech recognition systems, and they integrate these linear transforms into the maximum mutual information (MMI) estimation of HMM parameters.
•Proceedings Article
Discriminative linear transforms for feature normalization and speaker adaptation in HMM estimation.
Stavros Tsakalidis,Vlasios Doumpiotis,William Byrne +2 more
- 01 Jan 2002
TL;DR: In this paper, discriminative training procedures that employ linear transforms for feature normalization and for speaker adaptive training were introduced for improving large vocabulary conversational speech recognition systems, and they integrate these linear transforms into the maximum mutual information (MMI) estimation of HMM parameters.
36
•Proceedings Article
Lattice segmentation and minimum Bayes risk discriminative training
Vlasios Doumpiotis,Stavros Tsakalidis,William Byrne +2 more
- 16 Sep 2003
TL;DR: In this article, two approaches to incorporate discriminative training procedures in segmental minimum Bayes-Risk decoding (SMBR) are presented, which is used to segment lattices produced by a general automatic speech recognition system into sequences of separate decision problems involving small sets of confusable words.
Lattice segmentation and minimum Bayes risk discriminative training for large vocabulary continuous speech recognition
Vlasios Doumpiotis,William Byrne +1 more
TL;DR: Refinement of the search space that allows the use of specialized discriminative models is shown to be an improvement over rescoring with conventionally trained discrim inative models.
22
Pinched Lattice Minimum Bayes Risk Discriminative Training for Large Vocabulary Continuous Speech Recognition
Vlasios Doumpiotis,William Byrne +1 more
- 04 Oct 2004
TL;DR: Modelling techniques originally developed to improve search efficiency in Minimum Bayes Risk decoding can be used to transform these estimation algorithms so that exact update, risk minimization procedures can been used for complex recognition problems.