Search

22 May 2004: Factor analysed hidden Markov models for. speech recognition. A-V.I. Rosti, M.J.F. GalesCambridge University Engineering Department, Trumpington Street, Cambridge,. CB2 1PZ, UK. Abstract. Recently various techniques to improve the correlation model

22 May 2004: Togive an idea of the range of these N -best lists the oracle (best) er-ror rate on the tst data was 1.24% and the “idiot” (worst) errorrate was

29 Apr 2004: 29.44 0 6standard-BIC-judge, full cov 24.62 2 2standard-BIC-judge, diag cov 23.90 3 1standard-BIC-judge, 128mix GMM 23.76 5 1IDIAP-BIC-judge, full ... 37.92 10.74 24.62BIC-judge, diag cov 30.30 19.94 18.11 31.05 37.92 10.74 23.90BIC-judge, 128mix GMM

26 Jan 2004: The likelihood of an observation given thestate is simply. p(oj|x) = N(oj; Cx µ(o), Σ(o)) (24).

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

8 Oct 2004: ix. 2.7 Summary 23. 3 Generalised Linear Gaussian Models 24. 3.1 State Space Models 24. ... bank of triangular filters (eg. 24 channels). This smoothed power spectrum is compressed using.

23 Mar 2004: Eval03 with CU-HTK P1-P2 System. Overall Swbd Fisher Male Female. h5train03b LM03 24.6 28.7 20.2 25.7 23.5h5train03b LM03Fsh3896 23.9 28.2 19.3 ... 25.0 22.8fisher3896 LM03Fsh3896 23.1 27.0 18.9 24.6 21.6fisher3896h5 LM03Fsh3896 22.7 26.6 18.5 24.2 21.1.

13 May 2004: 92 32.06 24.95Num Poss 27 9 3 (39)Best 19.94 36.46 31.74 24.68. ... miss Worst 31.96 37.96 32.06 27.44allowed Mean 24.63 37.45 31.95 25.83.

25 May 2004: System MPE Prior eval03 Male Female. MPE-GI Dynamic MMI 25.9 27.3 24.5MPE-GD MPE-GI model 25.6 27.1 24.1. ... Gaussianised 29.8 21.9 26.0CN 28.7 21.3 25.1. CNC 28.1 20.8 24.6. •

28 Jul 2004: 4] M. Fischler and R. Bolles. Random sample consensus: A paradigm for model fitting with applications to image analysisand automated cartography.Graphics and Image Processing, 24(6):381–395, 1981.