Skip to main content

Table 2 Word recognition accuracy (%) of the Tucker 3-mode and probabilistic 2-mode based methods

From: Speaker adaptation in the maximum a posteriori framework based on the probabilistic 2-mode analysis of training models

   Number of Number of adaptation sentences
Method (K R ,K D ) free parameters 1 2 3 4 5
Tucker 3-mode (20, 35) 700 91.84 92.98 93.07 92.99 93.11
  (20, 38) 760 91.82 92.83 93.11 93.01 93.01
  (30, 35) 1050 90.77 92.99 93.18 93.09 92.94
  (30, 38) 1140 90.77 92.86 93.18 93.01 92.86
  (40, 35) 1400 89.39 92.85 93.11 93.24 93.03
  (40, 38) 1520 89.16 92.77 93.20 93.14 92.98
  (50, 35) 1750 87.95 92.34 93.24 93.26 93.13
  (50, 38) 1900 87.75 92.47 93.27 93.31 93.16
Probabilistic 2-mode (20, 35) 700 93.07 93.18 93.26 93.27 93.16
  (20, 38) 760 92.96 93.07 93.03 93.24 93.13
  (30, 35) 1050 92.98 93.20 93.24 93.24 93.31
  (30, 38) 1140 92.94 93.33 93.33 93.27 93.31
  (40, 35) 1400 93.13 93.20 93.39 93.24 93.01
  (40, 38) 1520 93.14 93.22 93.33 93.20 93.24
  (50, 35) 1750 93.26 93.35 93.37 93.29 93.29
  (50, 38) 1900 93.37 93.44 93.42 93.31 93.39
  1. The number of mixture components R=3472·8 and the dimension of acoustic feature vector D=39. The number of free parameters is K R ×K D .