Skip to main content

Table 3 English AMI meeting corpus perplexity and WER results using RNNLM and CUED-RNNLM against our proposed model

From: Enhancing recurrent neural network-based language models by word tokenization

 

PPL

WER

Dev

Eval

Dev

Eval

3g [25]

93.6

82.8

25.2

25.4

+CRNN (CE) [25]

83.3

75.2

23.9

24.1

+FRNN(CE) [25]

81.0

71.7

23.9

24.0

+FRNN(VR) [25]

80.4

71.6

23.9

23.9

+FRNN(NCE) [25]

81.1

72.8

24.0

24.1

+Proposed model

75.3

69.7

22.3

22.5