Automatic speech recognition using hidden Markov models

In a previous study [1] we presented a method for isolated-word speech recognition using a back-propagation network as a pattern classifier. A major limitation of using such a classifier is its inability to allow duration variability between different utterances of the same word. In this study, we present an algorithm for isolated-word recognition taking into consideration the duration variability. The algorithm is based on extracting acoustical features from the speech signal and using them as the input to Hidden Markov Models (HMMs) to recognize the spoken word. Fifty words, each uttered ten times, are tested for recognition. Our results show that our system was able to recognize these words.