Search

14 Sep 2010: These approaches either use the existing SVM parameters as a form of priorfor the samples from the new domain (Li and Bilmes, 2006), or define a resam-pling weight for the ... 6 Conclusions. This paper has described a new approach to noise-robust speech

26 May 2010: Previously we reported a. new method of sound-speed estimation in the context of image deconvolution. ... These new features are. found to be important in determining the speed in the dual-layer scenario.

25 Oct 2010: matching. In this paper, 1) a new discrim-inative method that maximises orthogonality between subspacesis proposed. ... The number of classesof the combined and the new set are denoted by.

25 Oct 2010: The learn-ing complexity over a new data set is not analytically pro-vided. ... step. The class posteriorprobabilities of the new samples are set to the probabilisticlabels.

25 Oct 2010: The learn-ing complexity over a new data set is not analytically pro-vided. ... step. The class posteriorprobabilities of the new samples are set to the probabilisticlabels.

10 Jun 2010: data, such as broadcast news and telephone speech recordings. Such found data often has. ... new approach for adaptive training based on discriminative mapping transforms is proposed.

25 Feb 2010: At each subsequent time instance t the HMM transitions into a new stateθt = j. ... An observation is then generated based on the output distribution, bj (ot) = p(ot|θt),associated with the new state.

6 Oct 2010: improving SMT quality by making it easier tointegrate new modelling approaches into good baseline systems.

20 Oct 2010: Figure 1.1: Graphical representation of noisy channel models and decoding. 1http://news.xinhuanet.com/english/2007-08/31/content_6637522.htm. ... It looks at the discriminative training of alignment modelsto improve the quality of word-level alignments