We construct an initial acoustic model from a combination of licensed corpora, an initial corpus of learner speech, and sometimes corpus data from similar languages and integrate it into the first beta version of the TLCTS course.
Earlier versions of TLCTS attempted to detect pronunciation errors on a continual basis in the Skill Builder (Mote et al.
One key challenge for the dialogues in TLCTS is to manage the variability of the spoken inputs a learner may produce.
As learners use TLCTS courses, it is important to track whether the learners are making progress toward learning objectives.
The TLCTS system developed over several years, starting as a research project at the University of Southern California (USC) Information Sciences Institute and later at Alelo.
Indeed, we believe that a good deal of the success of TLCTS has stemmed from acting upon the feedback from our users.
Just as we have restructured and improved the system architecture, we have progressively improved the curricula that TLCTS supports.
We have come to recognize the collaborative nature of content development in the TLCTS mode.
Some units combine TLCTS training with classroom training, whereas others rely exclusively on TLCTS for their language and culture training.
We continue to improve TLCTS based on experience gained with earlier versions of TLCTS courses.