Interactive Pronunciation Training System

Two different types of automatic speech recognizers will be evaluated for their effectiveness in an interactive Spanish-language pronunciation-training system. This effort is an extension of previous research on English language systems by this research group, and represents the first of several evaluations for other languages. One spectral-template based recognizer and two Hidden Markov Model (HMM) recognizers from different developers will be evaluated. The evaluation will benchmark recognizer accuracy in distinguishing between very similar Spanish words and will also determine the recognizers' capability to provide evaluative feedback of pronunciation accuracy in speech-training tasks. A prototype multimedia pronunciation-training system will be developed for native English speakers who are learning Spanish. This system will use the speech recognition technology determined to be most successful in distinguishing between intelligible and unitelligible Spanish. In Phase II a fully developed Spanish pronunciation training system with a wide variety of speech production and speech perception training drills will be produced. The same procedures will be applied to the design of modules for additional languages. These systems will have broad commercial and military applicability. Anticipated Benefits: Research and development on low-cost language trainers will contribute directly to efforts in the military to support multinational or multilingual operations. Wide availability of interactive, stand-alone language and pronunciation trainers will improve second-language proficiency and thereby enhance effective communication under adverse conditions. The potential commercial market for the proposed Spanish trainer in educational settings is very large.