Hindi Speech Recognition Using Hidden Markov Model Tool Kit

Abstract

Speech is most effective initial communication medium. It is also relates to production and perception. Speech is a medium of transferring information emotions and thoughts. Moreover it is also a unique medium of passing human intellect from one member to another. In science the human voice has long been an accepted feature.The purpose of this final Doctor of Philosophy degree research was to develop a Hindi speech recognition tool using HTK and to make the technology more accessible. The development includes an extensive study of Hidden Markov Model which is presently stated as the most popular tool in the field of speech recognition. The use of Hidden Markov Model Tool Kit HTK for speech recognition has become most predominant in the last several years as evidence by the number of published research papers at major conferences on speech. Database of isolated digits is prepared in clean and noisy environment. This research for acoustic features is explained with full description on Hindi speech recognition using HTK for isolated digits. In first phase of the research it is observed that Revised Perceptual Linear Prediction RPLP is giving best recognition efficiency amongst most of the conventional methods. In the past research pertaining to speech recognition was limited to the acoustic component of speech. Soundbased research has had many applications. This audiobased research gave rise to many landmark innovations.So there is in need to get Audio Visual features for better results of recognition. Audiovisual features play a very important role in ASR systems in presence of noise. In this thesis Hindi speech recognition system is designed using audio visual features also. In real life this speech recognition technology is very useful in traffic security or facilitates functional disability for people. This research work is provided better results for both Audio and Visual. newline newline newline newline newline newline