Skip to main content

Table 5 WER for evaluation set rescored with different RNNLMs and LM-KN5

From: RNN language model with word clustering and class-based output layer

 

Class

Model

50

100

200

400

LM-KN5 + RNNLM-Freq (%)

13.0

13.1

13.0

13.1

LM-KN5 + RNNLM-Brown (%)

12.7

12.7

12.7

12.4

  1. The WER of 1-best hypothesis is 14.1% and the WER for LM-KN5 is 13.8%.