Abstract: An optimal self-organization learning algorithm is presented in this paper. A new speech recognition system, which is based on the basis of SOM/HMM principles, bas been provided. In the system, the SOM networks are used as vector quantifier (VQ). After trained by the optimal self-organization algorithm, the SOM networks are adjusted by K-means clustering algorithm. experimental results showed that the new speech recognition method introduced in this paper can raise the recognition accuracy of the system.

Key words: speech recognition, neural networks, speech processing