 |
ISCA International Workshop on Speech and Language Technology in Education (SLaTE 2009)Wroxall Abbey Estate, Warwickshire, England |
 |
|
ISCA International Workshop on Speech and Language Technology in Education (SLaTE 2009)Wroxall Abbey Estate, Warwickshire, England |
|
Virtual Chinese Tutor (VCT) is a successfully operating online Chinese pronunciation
learning software,
specifically designed for giving the students opportunities to practice both their
listening and speaking
skills as many time as they wish anytime and anywhere. It was developed by a joint effort
between Academia
Sinica, National Taiwan University and some industry partners. The first version of VCT
has been completed
and made available on-line at http://chinese.ntu.edu.tw/. Virtual Chinese Tutor (VCT) is
able to evaluate
the pronunciation of each utterance produced by an individual learner from four different aspects:
pronunciation (i.e. the Initial (initial consonant) and Final (vowel or diphthong part
plus an optional
nasal ending) of each individual syllable (or character)), pitch (i.e. the lexical tone or
neutral tone of
each individual syllable (or character)), timing (i.e. the duration distribution among
different syllables
(or characters) in the utterance), and emphasis (i.e. the energy distribution among
different syllables
(or characters) in the utterance), as well as an overall score for the entire utterance. For those
phonemes with scores below a threshold, a 3-dimensional video will show on the screen to
demonstrate the
actions of the vocal tract shape, including the relative positions among the lip, tongue and other
articulators. After a learning session is finished, the diagnostic statistics for the learner is
automatically summarized, indicating the important directions for improvements. This
software platform can be
used with any course content as long as the text and voice files for the course content
are given. For each
utterance produced by the learner, forced alignment with the corresponding utterance
produced by the instructor
is first performed, and the pronunciation, pitch, timing and emphasis scores for each
syllable are then
evaluated by a set of acoustic models, tone models, and prosodic models. The overall
score is then the
weighted sum of all those scores. The scoring algorithm was trained with the scores given by real
professional Chinese teachers, over a corpus produced by a group of real learners whose
mother tongues
are not Chinese. Both the above training corpus and course content currently used in this
software were
contributed by the International Chinese Language Program of National Taiwan University.
The details of
the technologies will be explained when demonstrating the system in the technical program.
VCT is now also a part of the program offered by NTUtorMing
(http://www.ntutorming.com/), which is an online teaching institution that was jointly
established by
National Taiwan University (NTU) and TutorABC, a company located in Taipei focusing on language
education. Chinese Learners around the world can have real-time interaction with
professional Chinese
teachers through TutorABC’s online learning platform, and VCT offers after-class practice
or homework
for the students.
Bibliographic reference. Wang, Yow-Bang / Wang, Hsin-Min / Lee, Lin-Shan (2009): "Virtual Chinese tutor (VCT) - a Chinese language pronunciation learning software", In SLaTE-2009 (abstract).