Improving the accuracy of pronunciation lexicon using Naive Bayes classifier with character n-gram as feature: for language classified pronunciation lexicon generation

This paper looks at improving the accuracy of pronunciation lexicon for Malayalam by improving the quality of front end processing. Pronunciation lexicon is an in evitable component in speech research and speech applications like TTS and ASR. This paper details the work done to improve the accuracy of automatic pronunciation lexicon generator (APLG) with Naive Bayes classifier using character n-gram as feature. ngram is used to classify Malayalam native words (MLN) and Malayalam English words (MLE). Phonotactics which is unique for Malayalam is used as the feature for classification of MLE and MLN words. Native and nonnative Malayalam words are used for generating models for the same. Testing is done on different text input collected from news domain, where MLE frequency is high.