Prefix Tree based N-best list Re-scoring for Recurrent Neural Network Language Model used in Speech Recognition System

摘要

Recurrent Neural Network Language Model (RNNLM) has recently been shown to outperform N-gram Language Models (LM) as well as many other competing advanced LM techniques. However, the training and testing of RNNLM are very time-consuming, so in real-time recognition systems, RNNLM is usually used for re-scoring a limited size of n-best list. In this paper, issues of speeding up RNNLM are explored when RNNLMs are used to re-rank a large n-best list. A new n-best list re-scoring framework, Prefix Tree based N-best list Rescoring (PTNR), is proposed to completely get rid of the redundant computations which make re-scoring ineffective. At the same time, the bunch mode technique, widely used for speeding up the training of feed-forward neural network language model, is investigated to combine with PTNR to further improve the rescoring speed. Experimental results showed that our proposed re-scoring approach for RNNLM was much faster than the standard n-best list re-scoring~1. Take 1000-best as an example, our approach was almost 11 times faster than the standard n-best list re-scoring.