This paper describes a pronunciation error detection method for Repetitor, a pronunciation training computer program for second language learners of Dutch. A database of L2-speech was constructed and a selection of relevant pronunciation errors for Repetitor was made. Our error detection method is based on a weighted variant selection using forced alignment. Tested on the database, the results show that our method achieves satisfactory detection performance for most pronunciation errors yielding a precision of correct rejects of over 85% for most errors, and scoring accuracies between 85% and 100%.
Cite as: Sanders, E., Heuvel, H.v.d. (2010) Automatic pronunciation error detection in repetitor. Proc. Second Language Studies: Acquisition, Learning, Education and Technology (L2WS 2010), paper O2-4
@inproceedings{sanders10_l2ws,
author={Eric Sanders and Henk van den Heuvel},
title={{Automatic pronunciation error detection in repetitor}},
year=2010,
booktitle={Proc. Second Language Studies: Acquisition, Learning, Education and Technology (L2WS 2010)},
pages={paper O2-4}
}