Implementation of Discrete HMMs for Isolated Spoken Word Recognition

Isolated spoken words are acoustically separated from other spoken words by moments of silence. In this work, we have implemented Discrete Hidden Markov Models (DHMMs) for Isolated Spoken Words Recognition. Discrete HMMs have been successfully used in many temporal pattern recognition problems like speech recognition, handwriting recognition, gesture recognition, etc., for more than four decades. We have implemented a basic HMM from scratch in Java language and applied it to train and recognize isolated spoken words. We have tested our implementation in recognizing ten digits on a custom built speaker database. We have obtained an accuracy of 95 % and 82.5% for training and testing respectively.