Evaluation and implementation of a voice-activated dialing system with utterance verification

Abstract

This study investigates the utterance verification (UV) algorithm for a voice-activated dialing (VAD) system. The UV techniques help to improve the system accuracy of a VAD system and to improve the efficiency of user interface by reducing the need of confirmation. In this paper, we examine various UV methods, namely, all-phone garbage model (GM), N-best likelihood ratio (NBLR), and the combined methods. The performances of a VAD system with UV at various vocabulary sizes are studied. By rejecting 9.5% of correctly recognized names, the system error rate become less than 0.3% which represent a reduction of 91% in error rate over the baseline system. The UV technique can reduce the number of confirmation by at least 88% with a system error rate of 0.28%.

Extracted Key Phrases

8 Figures and Tables

Cite this paper

@inproceedings{Tan1998EvaluationAI, title={Evaluation and implementation of a voice-activated dialing system with utterance verification}, author={Beng Tiong Tan and Yong Gu and Trevor Thomas}, booktitle={ICSLP}, year={1998} }