Skip to main content

Advertisement

Table 2 Recognition performance with various hybrid lexicons

From: A hybrid input-type recurrent neural network for LVCSR language modeling

Tasks LM H-R1 H-R2
  #Word 30 K 25 K 20 K 30 K 25 K 20 K
  #PM 5 K 10 K 15 K 5 K 10 K 15 K
BN PER (%) 18.84 18.85 18.89 18.84 18.81 18.86
  WER (%) 22.16 22.19 22.24 22.15 22.13 22.18
VT PER (%) 18.79 18.86 18.76 18.41 18.41 18.56
  WER (%) 21.57 21.66 21.57 21.27 21.27 21.47
AVG. PER (%) 18.82 18.86 18.83 18.63 18.61 18.71
  WER (%) 21.87 21.93 21.91 21.71 21.70 21.83