In this paper

a basic principle called the equidistortion principle for vector clustering is theoretically derived by using Gersho’s asymptotic theory

and a new competitive learning algorithm is prorosed with a selection mechanism

called the CLS（Competitive and Selective Learning）algorithm.Because the selection mechanism enables the system to escape from local minima

the proposed algorithm can obtain better performance without a particular initialization procedure.A new neural network algorithm with competitive learning and multiple safe rejection schemes are proposed in the context of parallel

self organizing

hierarchical neural networks（PSHNN）.The input of PSHNN is a subset of the output scores of HMM.The experimental results indicate that the recognition ability of the method based on competitive learning neural network is higher than that of the traditional HMM method.