A study on the recognition of the isolated digits using recurrent neural predictive HMM

Author

Kim, Soo-Hoon ; Koh, Si-Young ; Ahn, Jeom-Young ; Hur, Kang-In

Author_Institution

Dept. of Electron. Eng., Dong-A Univ., Pusan, South Korea

Volume

1

fYear

1999

fDate

1999

Firstpage

593

Abstract

We composed the recurrent neural predictive HMM (RNPHMM) to provide the dynamic feature of the speech pattern for HMM. The RNPHMM is the hybrid network of the recurrent neural network and the HMM. In the experiment, we compared the recognition abilities of the RNPHMM as we increased the state number, prediction order, and number of hidden nodes for the isolated digits. The models of the recurrent neural predictive HMM are the Elman network prediction HMM and the Jordan network prediction HMM. As a result of the experiments, the Elman network prediction HMM and the Jordan network prediction HMM have a good recognition ability of 98.5% for test data respectively

Keywords

feature extraction; hidden Markov models; prediction theory; recurrent neural nets; speech recognition; Elman network prediction HMM; Jordan network prediction HMM; dynamic feature; experiment; experiments; hidden Markov model; hidden nodes; hybrid network; isolated digits recognition; prediction order; recurrent neural network; recurrent neural predictive HMM; speech pattern; speech recognition; test data; Covariance matrix; Hidden Markov models; Neural networks; Pattern recognition; Predictive models; Probability; Recurrent neural networks; Speech recognition; Statistics; Testing;

fLanguage

English

Publisher

ieee

Conference_Titel

TENCON 99. Proceedings of the IEEE Region 10 Conference

Conference_Location

Cheju Island

Print_ISBN

0-7803-5739-6

Type

conf

DOI

10.1109/TENCON.1999.818484

Filename

818484