This paper describes the systems developed by FBK for the 2019 Spoken CALL Shared Task, that requires to automatically grade Swiss students, speaking German, that have to answer in English to German prompts. All answers are automatically transcribed, using an Automatic Speech Recognition (ASR) system, and labelled as accept or reject by a classifier. We developed an improved version of the baseline ASR system (made available by the organizers of the challenge), that has been used to produce better automatic transcriptions, from which a set of linguistic features are derived. Then, features vectors, computed at sentence level, are fed into a neural network based classifier that predicts the labels. In this paper we describe the details of the developed ASR system, as well as the set of features used in the accept/reject classification task. We also discuss the impact of subsets of features on the final classification performance.
Cite as: Gretter, R., Matassoni, M., Falavigna, D. (2019) The FBK system for the 2019 Spoken CALL Shared Task. Proc. 8th ISCA Workshop on Speech and Language Technology in Education (SLaTE 2019), 6-10, doi: 10.21437/SLaTE.2019-2
@inproceedings{gretter19_slate,
author={Roberto Gretter and Marco Matassoni and Daniele Falavigna},
title={{The FBK system for the 2019 Spoken CALL Shared Task}},
year=2019,
booktitle={Proc. 8th ISCA Workshop on Speech and Language Technology in Education (SLaTE 2019)},
pages={6--10},
doi={10.21437/SLaTE.2019-2}
}