Liulishuo's System for the Spoken CALL Shared Task 2018

The Spoken CALL (Computer-Assisted Language Learning) 2018 shared task requires systems to automatically accept or reject each single-sentence spoken response depending on whether the response is correct given a prompt. Spoken responses are first recognized into texts and then classified as ‘accept’ or ‘reject’ based on their language and meaning. This paper describes our system for the shared task. We focused on improving speech recognition performance, developing a rich set of features to capture the linguistic and semantic meaning of the responses and optimizing classification results for various factors (training set, n-best hypotheses of speech recognition, decision threshold, model ensemble). Our system achieves the best performance among the participating teams.