Continuous Sinhala Speech Recognizer

Abstract

Speech is the most natural way of communication among humans. Speech recognition is the process of transforming a speech signal into its corresponding word sequence. When the recognition is carried out by a computer program, it is known as Automatic Speech Recognition (ASR). Nowadays the use of speech recognition systems is very popular because it directs a friendlier interface between human and computer. The day which we will be able to have computers that understand our voice commands and our behavior to assists us is not far away. Speech applications to recognize English and some other major languages have already been developed and currently most of them are in a much matured stage. However, there are no such applications for Sinhala speech recognition. Although there are few researches carried out to implement discreet Sinhala speech recognizers, the research on building a continuous Sinhala speech recognizer is very poor. This thesis is based on an attempt to build an Automatic Speech Recognition system for continuous Sinhala speech. The underlying concept of building the recognition system is Hidden Markov Modeling (HMM). The system is built using the HTK tool kit (Hidden Markov Model Toolkit). The tests carried out for the build system shows a considerable performance for this primary stage.