Skip to main content

Table 11 LW-BGRU and LW-BrGRU accuracy as a function of the number of layer and hidden layer size

From: Advanced recurrent network-based hybrid acoustic models for low resource speech recognition

# layers Model # hidden layer size  
   300 500 700 800 900
2 A 48.4 47.6 47.4 47.4 48
  B 48.1 47.3 47 47.1 47.8
  C 43.9 43.3 43.1 43.2 44.1
  D 43.7 43.1 42.7 42.9 43.5
3 A 48.2 47.7 47.9 48.5 48.9
  B 47.5 46.9 46.9 47.1 47.6
  C 43.8 43.5 43.7 44.3 44.9
  D 42.9 42.6 42.6 43.1 43.5
  1. Here, “A”, “B”, “C,” and “D” represent “W-BGRU-fbank,” “LW-BrGRU-fbank,” “LW-BGRU-MBN,” and “LW-BrGRU-MBN,” respectively