Durational Constraints for Network Based Connected Digit Recognition

Authors

  • Marcia A. Bush

Abstract

This paper examines the influence of durational constraints on recognition accuracy in an acoustic-phonetically based, speaker-independent connected digit recognizer. The constraints arc expressed using a set of finite-state pronunciation networks, together with specifications of minimum and maximum allowable durations for network primitives. The recognizer was tested on a corpus of 1232 5-digit and 7-digit strings, with and without a priori knowledge of string length. Recognition accuracies ranged from 33.9% to 94.6% and from 91.6% to 96.8%, for unknown and known string lengths, respectively, depending on the particular durational constraints incorporated in the network models.

Additional Files

Published

2022-12-03

How to Cite

1.
Bush MA. Durational Constraints for Network Based Connected Digit Recognition. Canadian Acoustics [Internet]. 2022 Dec. 3 [cited 2024 Nov. 21];14(3 bis):89-90. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/3534

Issue

Section

Proceedings of the Acoustics Week in Canada