A simple tamil speech recognition system based on cmusphinx

Abstract

This paper focuses on the research of phonemes and their usage for Tamil speech recognition using the CMUSphinx API [4]. Tamil did not have a solid speech recognition application especially the Tamil that is spoken on a daily basis. This paper is the outcome of the authors building a new dictionary for 418 words (as of April 27 2019). This paper is intended solely to show the result of mapping Tamil grapheme to phoneme in hope that it will help similar sounding languages and the development of the Tamil language itself. The model used recorded speech to map the graphemes in the built dictionary to the phonemes in the recorded words. The accuracy of the model is represented by determining the Word Error Rate generated by the language model.