5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

Selection of the Optimal Structure of the Continuous HMM Using the Genetic Algorithm

Tomio Takara, Yasushi Iha, Itaru Nagayama

University of the Ryukyus, Japan

The hidden Markov models (HMMs) are widely used for automatic speech recognition because they have a powerful algorithm used in estimating the model's parameters, and also achieve a high performance. Once a structure of the model is given, the model's parameters are obtained auto- matically by feeding training data. However, there is still an unresolved problem with the HMM, i.e. how to design an optimal HMM structure. In answer to this problem, we proposed the application of a genetic algorithm (GA) to search out such an optimal structure, and we showed this method to be effective for isolated word recognition. However, the test of this method was restricted to discrete HMMs. In this paper, we propose a new application of the GA to the continuous HMM (CHMM) which is thought to be more effective than the discrete HMM. We report the results of our experiment showing the effectiveness of the genetic algorithm in automatic speech recognition.

Full Paper

Bibliographic reference.  Takara, Tomio / Iha, Yasushi / Nagayama, Itaru (1998): "Selection of the optimal structure of the continuous HMM using the genetic algorithm", In ICSLP-1998, paper 1066.