Study of Hidden Markov Model for Isolated Word Recognition

Abstract: Speech processing is a very wide field and Speech recognition covers big part of it. It is a measure of converting an acoustic signal taken from microphone to a set of words. Speech recognition is a very important aspect for voice driven service portals, speech interface in automotive navigation and guidance system or speech driven application, and isolated word recognition is a base of speech recognition. In this project, the isolated word recognition system is designed which is based on Hidden Markov Model. The recognition process is divided into four stages. It starts with Analysis followed by Feature extraction. In feature extraction important features are extracted such as pitch, pitch duration etc. In next stage acoustic model is formed and recognition stages comes followed by acoustic model. Recognition process makes use of HMM with 3 hidden states. HMM used here is context-dependent that is monophone based. In HMM three algorithms are used, forward algorithms for finding out probability distribution, Viterbi algorithm is used to find hidden states and Baum-Welch algorithm is used to fit the model. Word is recognized with higher accuracy at last stage.

Keywords: Speech recognition, Hidden Markov Model, Acoustic model.