ISCA Archive SLaTE 2017
ISCA Archive SLaTE 2017

The CSU-K Rule-Based Pipeline System for Spoken CALL Shared Task

Nico Axtmann, Carolina Mehmet, Kay Berkling

This paper presents the set-up and results regarding the Cooperative State University submitted system for the shared spoken CALL ESL task. The data was collected from Swiss teenage students using a speech-enabled online tool for English conversation practice. The tasks consisted of training data of a German text prompt with the associated audio file containing an English language response by the students. Problems therefore consisted of recognizing children's speech with foreign accent, grammatical and vocabulary problems and a number of false starts due to language learning issues. The task is to create software that will decide whether each response is appropriate (accept) or inappropriate (reject) in the context of the prompt. Results are reported through a single D-value that is computed specifically for this task. The contribution of this paper is a detailed analysis of a va- riety of changes to the baseline system and the analysis of their contribution to the overall performance. The paper reports the official result on the shared task test files but also goes beyond the originally submitted system.