Application of Modified Gravitational Search Algorithm to Solve the Problem of Teaching Hidden Markov Model

Hidden Markov Model is a finite series of states that is continues with a probability distribution in a special state, an output can be obtained by continuous probability distribution. Since states are hidden from outside, this model is called Hidden Markov Model. In ordinary Markov Model, the position is directly visible to observer so probabilities transference state will be the only parameters. In Hidden Markov Model, the position is not visible directly but the affected variants by the position are visible. Each state taken for a possible output will have a probability distribution. Therefore, the sequence of taken states created by HMM would provide some information about the sequence state. Hidden Markov Models will be distinguished for their instruction in identifying the temporary patterns such as speech, handwriting, identifying hint and pointing, bioinformatics and so on. In this paper, a new method based on Modified Gravitational Search Algorithm (MGSA) has been used to improve the teaching of Hidden Markov Model (HMM). The teaching of HMM is based on Baum-Welch algorithm (BW). One of the problems of HMM teaching is the absence of any assurance about reaching of this algorithm to global optimum and the convergence of this method is often towards a local optimum. In this paper, the Modified Gravitational Search Algorithm has been used to exit Baum-Welch from local optimum and search for other optimal points. Furthermore, we have compared the proposed algorithm with two algorithms, PSO and Ant Colony, which have been used finally in Speech Recognition.