2013 Fiscal Year Research-status Report
ジェスチャ及び音声認識を用いたコミュニケーション能力育成教材の開発と実践評価
| Project/Area Number |
25370681
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Research Institution | National Institute of Technology, Toyama College |
|---|
Principal Investigator |
COOPER T・D 富山高等専門学校, 一般教養科, 准教授 (70442449)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
塚田 章 富山高等専門学校, 電子情報工学科, 教授 (40236849)
的場 隆一 富山高等専門学校, 電子情報工学科, 助教 (30592323)
成瀬 喜則 富山高等専門学校, 国際ビジネス学科, 教授 (00249773)
|
|---|
| Project Period (FY) |
2013-04-01 – 2016-03-31
|
| Keywords | communication / gesture recognition / facial recognition |
|---|
| Research Abstract |
We are developing a software-based gestural [GR], facial [FR] and speech recognition [SR] system to help students perform better on English invterviews and presentations. GR evaluates users’ gestures on quality and quantity. SR automates the system giving instant feedback. FR guarantees a quick, secure and authenticated login. The project comprises 2 areas [system and databases], divided into 3 phases, with increases in test subjects, question difficulty and database size. The audio and video databases allow EFL researchers to identify common errors, assess current curricula and improve language and cognitive development. Our plan called for the modification and design of software / system prototype. First, a database of 10-15 questions and gestures for small-group study. Then increase to 50 questions: 30 basic / 20 intermediate and begin categorization. Finally increase the database to 100 questions 40 basic / 40 intermediate / 20 advanced and complete categorization. Finally, we are to present our findings at both domestic and international conferences.
2013: We created the 15 questions for the interview and carried out some tests of the SR software. We are experimenting with different SR engines and modified databases using the Sphinx SR engine. We are continuing to improve the accuracy. Through use of a 3-D camera we increased the FR accuracy and reduced the negative effect of other variables such as lighting, distance, etc. The GR component now includes a Hidden Markov Model which allows detection of key shapes and has increased the gesture accuracy.
|
| Current Status of Research Progress |
Current Status of Research Progress
3: Progress in research has been slightly delayed.
Reason
We experienced a slight delay due to the fact that our research is strongly linked with emerging fields in both technology and psychology. We have not reached the accuracy levels yet for with our customized SR database. We need to collect more data for this. We are currently experimenting with new cameras with higher depth resolution which will improve the results of our final findings. The gestural databases can be only created after collecting many hours of filming student presentations and speeches. We have been compiling these videos for future analysis.
|
| Strategy for Future Research Activity |
Our plan for the current year is to improve the function of the GR, SR, and FR components of our system. Testing will be carried out in large groups of 40. We will increase our question database to 50 questions and begin the categorization of the questions. 10 gestural patterns will be created and tested. Our SR database will increase, thereby improving accuracy. We will also test the accuracy of several mics, as well as the use of Jawbone technology to improve the quality of our system in large classrooms with lots of background noise. Our findings will be presented to the public at at least 1 national and 1 international conference.
|
| Expenditure Plans for the Next FY Research Funding |
We had planned to present several topics of our research at an overseas conference together in a small group. Due to the small delay in our research we decided to postpone the excursion to the following year.
Our plan is to use the money in the same way, to fund a research trip for two reserachers to present at a conference in the current academic year [2014-2015].
|
