A method for efficient DP matching in spoken word recognition

Author

OHNO, Sumio ; Fujisaki, Hiroya ; Hirose, Keikichi

Author_Institution

Dept. of Appl. Electron., Sci. Univ. of Tokyo, Japan

fYear

1994

fDate

13-16 Apr 1994

Firstpage

714

Abstract

A method has been developed for increasing the computational efficiency in spoken word recognition based on template matching using dynamic programming (DP matching). It is based on the detection and utilization of points along the time axis where the acoustic features of the speech signal are quasistationary. These quasi-stationary points (QS-points) are determined both for the input speech and for the stored templates, and are used for (1) pre-selection of candidate templates for DP matching, and (2) reduction of the search space for piecewise DP matching between corresponding intervals of the input signal and a candidate template. A recognition experiment on a 50-word vocabulary indicates that the method can reduce the computation time to 36% of the time required for a conventional DP matching method

Keywords

dynamic programming; speech recognition; acoustic features; computation time; computational efficiency; dynamic programming; efficient DP matching; quasi-stationary points; search space; spoken word recognition; template matching; time axis; Acoustic signal detection; Computational efficiency; Hidden Markov models; Image processing; Image recognition; Intelligent networks; Neural networks; Space stations; Speech processing; Speech recognition;

fLanguage

English

Publisher

ieee

Conference_Titel

Speech, Image Processing and Neural Networks, 1994. Proceedings, ISSIPNN '94., 1994 International Symposium on

Print_ISBN

0-7803-1865-X

Type

conf

DOI

10.1109/SIPNN.1994.344812

Filename

344812