Myanmar-English Code-Switched Automatic Speech Recognition System

This paper describes a Myanmar-English CodeSwitched Speech Recognition for Bilingual Information Technology(IT) lecture speech. The proposed recognizer is developed using the classical Gaussian Mixture Model-based Hidden Markov Model and Deep Neural Network(p-norm) activation function approach, trained on the Myanmar-English Code-Switched Speech Dataset(MEASR).MEASR dataset focuses on spontaneous speech containing Myanmar-English intra-sentential code-switching, collected from real online IT teaching sessions. In order to support both languages in the recognition process, two pronunciation dictionaries are used: the MyanmarDict dictionary and the CMU dictionary. We investigate the performance of the Myanmar-English code-switched IT lecture speech recognition. Experimental results reveal that the proposed system performs a best Word Error Rate of 22.23%, which proves its efficiency regarding bilingual and codeswitched speech scenarios.