A Pilot Study of the Relationship Between Automatic Analytic Scoring and Speech Accuracy: Implications from Log Data on Speech-Training Courseware

ID: 54597 Type: Brief Paper
  1. Yuichi Ono, Takeshi Kato, Shoma Aota, and Naoki Takebayashi, University of Tsukuba, Japan
  2. Yuki Toyono, CHIeru Co., Ltd., Japan

Wednesday, March 20 3:00-3:20 PM Location: Sunset 4 View on map

No presider for this session.

Abstract: This paper attempts to show how factors of the phonetic score, collected by Automatic Speech Recognition (ASR), relate the overall evaluation of speech in the context of Japanese English as a Foreign Language (EFL) learners working with online pronunciation-training courseware. The analytic scores are “pronunciation,” “intonation,” “accent,” “timing,” and “proficiency.” We conducted a multi-level correlation analysis and the result was that the total score was mainly correlated with the “pronunciation” variable for both sentence practice and word practice. Other findings are that the total score had a higher correlation with “intonation” for sentence practice, and that the “timing” variable was negatively correlated with the total scores for word practice. These findings provide hints to foreign-language instructors on how to integrate ASR-based courseware into a course curriculum for foreign-language teaching.


Conference attendees are able to comment on papers, view the full text and slides, and attend live presentations. If you are an attendee, please login to get full access.