Grammatical-based Word Segmentation Algorithm for Building Language Model in Myanmar ASR

Abstract

In this paper a grammatical-based word segmentation algorithm is presented. As the nature of Myanmar language, word segmentation process is necessary for machine translation, speech recognition, speech synthesis, etc. Many segmentation methods for Myanmar have been proposed for many reasons. Language model is the main part of computing on statistical speech recognition method. This word segmentation algorithm is proposed to use in building word-based language model for continuous speech recognition. We have used 16950 sentences for word segmentation. In this algorithm, we used post-positional marker and conjunction as word token markers. The output word lists are used to compute the probability of word order sequence for speech recognition system.