Word recognition based on the combination of a sequential neural network and the GPDM discriminative training algorithm

Abstract

The authors propose an isolated-word recognition method based on the combination of a sequential neural network and a discriminative training algorithm using the Generalized Probabilistic Descent Method (GPDM). The sequential neural network deals with the temporal variation of speech by dynamic programming, and the GPDM discriminative training algorithm is used to discriminate easily confused words by enhancing the distinguishing sounds of them during the scoring procedure. A Mandarin digit database uttered by 100 speakers was used to evaluate the performance of this method. The recognition rates are 99.1% on training data and 96.3% on testing data.<>