Conversion of NNLM to Back-off language model in ASR

Abstract

In daily life, automatic speech recognition is one of the aspect which is widely used for security system. To convert speech into text using neural network, Language model is one of the block on which efficiency of speech recognition depends. In this paper we developed an algorithm to convert Neural Network Language model (NNLM) to Back-off language model for more efficient decoding. For large vocabulary system this conversion gives more efficient result. Efficiency of language model depends on perplexity and Word Error Rate (WER)