We have designed a Turkish dictation system for Radiology applications. Turkish is an agglutinative language with free word order. These characteristics of the language result in the vocabulary explosion and the complexity of the N-gram language models in speech recognition. In order to alleviate this problem, we propose a task-specific, radiology, dictation system. Using words as recognition units, we achieve 87.06 % recognition performance with a small vocabulary size in a speaker independent system. The most common reason of errors during the recognition is due to the pronunciation variations across speakers, and also due to the inaccurate modeling. In this paper, to overcome these problems, we proposed a pronunciation modeling technique in which variation is modeled at the lexicon level. The pronunciation variants are selected by learning the common mistakes of our speech recognition system. As a proof of the concept, firstly we apply this method to the isolated recognition of small vocabulary radiological words. Our preliminary results show that, 24.74 % error rate reduction can be achieved for isolated word recognition. This idea can also be generalized to continuous speech recognition problem with a moderate vocabulary size.
Cite as: Arisoy, E., Arslan, L.M. (2004) Turkish radiology dictation system. Proc. 9th Conference on Speech and Computer (SPECOM 2004), 310-314
@inproceedings{arisoy04_specom,
author={Ebru Arisoy and Levent M. Arslan},
title={{Turkish radiology dictation system}},
year=2004,
booktitle={Proc. 9th Conference on Speech and Computer (SPECOM 2004)},
pages={310--314}
}