Automatic speechreading using genetic hybridization of Hidden Markov Models

In this paper, we present an Automatic SpeechReading system (ASR) that uses for modeling the visual speech an Hidden Markov Model (HMM) who is optimized by a Genetic Algorithm (GA). The idea is to combine a GA to explore the whole solution space with Baum-Welch algorithm in the training process to find the values of the exact parameters of the optimum. The experimental results show that the GA/HMM achieved higher rate recognition with less computation compared to the traditional HMM.