An improved multisection vector quantization model with application to Chinese digits recognition

In this paper, an improved model using multisection vector quantization (MVQ) is proposed and detailed experimental results are discussed. The major weakness of conventional MVQ is that it can not make segments accurately. Thus, a dynamic segmenting algorithm based on the Viterbi algorithm is presented. Moreover explicit state duration density and dynamic features are integrated into the model. Thus more acoustic information is depicted and a better performance is acquired