Search

14 Jun 2004: DARPA Broadcast News Recognition Workshops. Last modified: Fri Oct 11 19:55:20 BST 2002.

29 Apr 2004: Appendix - Cluster Voting Results by Show. System ABC VOA PRI NBC CNN MNB TOTALCUED-diary-bic (input 1) 32.03 20.78 21.40 32.06 37.92 10.74 ... 12CUED-diary-cost (input 2) 29.26 19.82 20.48 31.56 37.18 29.34 27.09Best CVOS score 26.71 18.43 18.11 29.84 37

14 Jun 2004: 43.60. CU. PER. 0.4. 0.2. 9.1. 11.60. 29.7. 32.20. MIT. CU. ... MIT. MIT. 2.3. 3.7. 5.6. 25.93. 20.6. 40.96. MIT. PER. 0.6.

25 May 2004: Gaussianisation. • Transform any distribution to standard Gaussian N(0, I). 20 15 10 5 0 5 10 15 200. ... Gaussianised 29.8 21.9 26.0CN 28.7 21.3 25.1. CNC 28.1 20.8 24.6. •

29 May 2004: Gauss 24.0 20.7 20.7 17.7 16.0WER (%) 35.3 35.1 35.2 35.3 35.5. ... Using the “opti-mal” structure, determined with α = 0 and 4 iterations of structureoptimization the error rate fell to 35.1% and the average numberof components per state was 20.7.

9 Dec 2004: Frames from phrase:SHOW THE GRIDLEY’S. Legend. • True• HMM• SLDS. 20 40 60 80 10035. ... Cambridge UniversityEngineering Department. Institute of Statistical Mathematics 20. SVMs, Generative Kernels and Maximum Margin Statistical Models.

23 Mar 2004: Cluster-based mean and variance normalisation was used. • ML search for warp-factor done using parabolic search in the range [0.80,1.20]. • ... 20% parameters reduction. • 0.5%0.7% abs improvement over global HLDA system. •

25 May 2004: 0 5 10 15 20 25 3026. 26.5. 27. 27.5. 28. ... 0 5 10 15 20 2526. 26.5. 27. 27.5. 28. 28.5.

23 Dec 2004: ldc04swm03S2 12 36.3 48.2S3 16 36.1 47.9S4 20 36.0 47.2. Table 4. ... The use of the GWD gave significant reductions in theperplexity. On the eval04 data an almost 20 point reduction in per-plexity was obtained.

29 Apr 2004: 8#PFMs 10 td 70-30 28.8 29.1 18.5 19.5 47.3 48.6#PFMs 20 td 50-50 33.2 34.8 13.9 14.8 ... The DB98 SULMs reduce the DEL error by c.20% abs (while increasing the INSerror by c.20% abs) compared to the LDC SULMs.

21 Sep 2004: Fortunately a number of efficient approximate algorithms have been proposed suchas graph cuts [1] and belief propagation [20]. ... Szeliski. Layered depth images.Computer Graphics,32(Annual Conference Series):231–242, 1998. [20] J. Sun, H,-Y Shum, and N

23 Dec 2004: State-of-the-art systems still yield high word error rates typicallywithin a range of 20–30%. ... global MLLR 20.2 35.8 36.4 30.7MPE - 19.3 35.6 35.8 30.1.

25 May 2004: 16 20 24 28. dev03 bnac 144 17.9bnacTDT4 375 16.8bnacTDT4TDT2 749 16.8 16.5 16.4bnacTDT4TDT2TDT4a 1126 16.7 16.3 16.0 16.1bnacTDT4TDT4a 752 16.7

4 Mar 2004: be estimated [20], whichremoves one degree of freedom in the ambiguities of the reconstruction. ... 12] M. Zerroug, R. Nevatia, Volumetric descriptions from a single intensity image, Int.Journal of Computer Vision 20 (1/2) (1996) 11–42.

25 May 2004: The testing adaptation isperformed on 1994 NAB Spoke 3 (s3-dev and s3-eval) task withan enrollment set (40 utterances) and a testing set (about 20 utter-ances) for each speaker.

18 Jun 2004: Bjik :=. ek ei 〈ej,V〉 (20). as boundary integral constants with V denoting the outward normal along. ... IEEE Trans. Image Processing 7 (1998) 398–410. 20. Schwarz, H.: Numerische Mathematik.

27 May 2004: CNC P4.[ACD]P5.[ACD] 24.1 17.1 20.7. Table 6. %WER on eval03 for all stages of 2003 system.

26 Jan 2004: 19). Σ̂(o). jm =1. PT. t=1 γ(o)jm(t). TX. t=1. γ(o)jm(t)diag. “. oto′. t (20). h. Ĉj µ̂(o)jm. i h.

11 Feb 2004: Next, control points are initialised at regular intervals(20 pixels) along every model edge. ... In Proc. IEEEIWAR’98, San Francisco, November 1998. [20] H. Rehbinder and B.

26 Nov 2004: 3 METHOD 9. 5 10 15 20 25 30 35 40 45 5020. ... Multiparametric attenuation andbackscatter images for characterization of carotid plaque. Ultrasonic Imaging 22 (1), 20–34.