This Small Business Innovation Research Phase I project will explore the technology needed to allow the development of voice-interactive computer aids for learning spoken languages. The project will make use of state-of-the-art speech recognition and understanding technology developed at MIT as part of the ARPA speech program. Specifically, the proposed work will use statistical segmental models to distinguish between correct and incorrect pronunciations by students learning a second language. These models will be trained in a similar way to the models used for speaker-independent speech recognition. In the Phase I research project, the feasibility of this approach will be shown by collecting a pilot speech corpus of students speaking the target language, the corpus will be labeled according to human judgments of pronunciation, and statistical models will be trained and tested on separate portions of this corpus.
