Search

He has authored and co-authored more than 200 papers.

27 Jul 2020: Abstract for rosti_csl04. Computer Speech and Language, 18(2):181-200, 2004. FACTOR ANALYSED HIDDEN MARKOV MODELS FOR SPEECH RECOGNITION.

23 Feb 2002: Proc. TREC-7, pp. 191-200 (Gaithersburg, Maryland, 1998). 6. S.E. Johnson [Reducing Word Error Rates of Found Speech: XPERT Tool for Transcription Analysis.

17 Nov 2017: Eval{1,2,3} sets (about 13 hours) contain spontaneous speech from 200 speakers with Gujarati, LA Spanish and mixed L1s, respectively. ... 5. Parse Tree. 7. Conclusion. Trn3 set contains trn1 and 200 hours selected from the unsupervised set.

27 Jul 2020: Semi-automatic image processing facilitates alignment of the beam and wires to within a tolerance of around 200 microns, despite the considerable beam thickness.

17 Nov 2017: c) (d). 0 50 100 150Frame Index. 0. 100. 200. 300. ... 100. 200. 300. 400. Frequency. (H. z). k 0 = 3 k 0 = 5 k 0 = 7.

28 Apr 2005: Error Analysis. c.200 DEL errors analysed by hand:. • 40.7 %: before asyndetic clause boundaryEx: they destroyed all the national monuments () he destroyed a large area. • ... c.200 INS errors analysed by hand:. • 26.2 %: before potential

17 Nov 2017: 403. 404. 50. 100. 150. 200. 250. 300. 350. 3762 7. ... I Cumulative state data usage statistics for 3-hour configuration (VLLP). 100 200 300 400 5000.

30 Jan 2000: Images 1 and 5200 0 200. 400. 200. 0. 200. Images 2 and 3. ... 200. Images 4 and 5. Figure8: Imagesof the ellipsesandepipolarlines after convergence.Eachfigure showsthe overlappingof images and , for -GP& 555 # and ¤@@P 555.

10 Dec 2003: 0. 0.5. 1. 1.5. 2. 0 200 400 600 800 1000. % ... 0. 0.2. 0.4. 0.6. 0.8. 1. 1.2. 1.4. 1.6. 0 200 400 600 800 1000.

17 Nov 2017: BBC Monitoring 200 journalists monitoring global audio, video, text news. Deutsche Welle Multilingual content generation and management.

8 Mar 2000: 100 samples were used in the Monte Carlo simulations. Table 1 shows the average one-step-ahead prediction errors obtained for 10 runs, of 200 time steps each, on a Silicon

23 Mar 2004: Key Issues: 2004 Data. • New data planned for collection in 2004 includes– more Fisher English CTS data– 200 hours of Levantine Arabic CTS data– 200 hours of Mandarin CTS data– English

3 Jul 2018: Results. # of Clusters 50 100 150 170 200 TextTiling [Hearst 1997] 0.484 DNN-HMM [Yu et.

14 Aug 2000: BRF is run on the parallel collection adding the best 200 new terms taken from the top 10 documents to the original document with a term frequency equal to 1.

25 May 2004: Gaussianisation. • Transform any distribution to standard Gaussian N(0, I). 20 15 10 5 0 5 10 15 200. ... 0.01. 0.02. 0.03. 0.04. 0.05. 0.06. 0.07. 0.08. 0.09. 0.1. 20 15 10 5 0 5 10 15 200.

20 Nov 2017: Linear Conjugate Gradient. Experimental Setup. I Experiments used ASRU 2015 MGB challenge data.I Training set : 200 hr MGB1 training dataset.I Validation : the official MGB1 dev.sub set (5.5

23 Feb 2002: s1. -. -. 12. 26. 59.11. 57.14. 54.04. 50.65. s1. 200. ... 10. -. -. 50.76. 49.42. 52.91. 51.67. s1. 200. 10. 8.

19 Jan 2018: rate thanthe 200 frames-per-second typical in many sys-tems.

7 Oct 2009: Frame accuracies are given foreachof the 200 hours subset MLPs in Table 1 and all subset networksgive similar results. ... 1The best performing 200 hours subsets MLPs.2The GALEeval07 non-sequestered testset version is used.