Enhancing spoken connected-digit recognition accuracy by error correction codes - a novel scheme

Abstract

Recognizing spoken connected-digit numbers accurately is an important problem and has very many applications. Though state-of-the-art word recognition systems have gained acceptable accuracy levels, the accuracy of recognition of current connected spoken digits (and other short words) is very poor. In this paper, we develop a novel scheme to enhance the accuracy of recognizing a connected number. The basic idea proposed in this paper is to increase the number of digits in a number and use these appended digits to increase the overall accuracy of recognizing the number, as is done in the error-correcting code literature. We further show that the developed scheme is able to uniquely and exactly correct single-digit errors.