A novel speech recognition approach based on multiple modeling by hidden Markov models

Most conventional automatic speech recognition systems are based on Hidden Markov Models (HMMs). Independently of the application it is well known that the Expectation-Maximization (EM) algorithm, commonly used to estimate HMM parameters is sensitive to the values of the initial model, and guarantees only a locally optimal solution. In this paper, we propose a novel approach for speech recognition based on multiple modeling by HMMs. Unlike traditional markovian approach where only one model is associated to every word, in our approach several models coming from different random initializations are combined for each word. The choice of these models is done by Genetic Algorithm (GA). Experimental results on one Arabic digit numbers dataset show the effectiveness of our approach by comparing against classical one based on single modeling (increasing the recognition accuracy of more than 90%).