Speech Recognition System For North-East Indian Accent

Abstract

Speech recognition is the process of converting an acoustic wave-form into the text containing the similar information conveyed by the speaker. This paper presents a speech recognition system for English digits in Indian (especially North Eastern) accent. Hidden Markov Model Tool kit (HTK-3.4.1) is chosen to implement the Hidden Markov Model as classifier with several set of Hidden Markov Model mixture. Mel Frequency Cepstral Coefficients are used as speech features. Experiments were performed for data collected in natural noise environment. The performance is evaluated using recognition rate. Hidden Markov Model state numbers and number of mixtures are investigated and possible directions for future research work are suggested.