This Small Business Innovation Research Phase 1 project will develop automatic pitch error detection (pinpointing) for non-native speakers and determine whether software using this detection method helps people improve their pronunciation. Over 50 million people worldwide spend over $50 billion per year learning to speak English. Carnegie Speech's NativeAccent teaches English sounds and rhythm, but not pitch since its detection was unreliable in the past. New detection methods, however, will enable the software to show differences between non-native and native speakers' utterances. Correct pitch makes speech understandable over more than one sentence as the listener is guided from one high-content word to the next. Recent cognitive science findings have produced successful techniques for training non-native speakers to contrast phone sounds. This project will adapt those techniques to pitch contrast, specifically, tones in Mandarin. Tonal training breaks the pitch problem down into one basic type variation, later building up to larger, more complex pitch contours and applying the techniques to English. Trainees will use a game with feedback piloted by NativeAccent pinpointing and intelligent tutoring techniques. Success in Phase 1 will be measured by whether people can be taught to produce tones more effectively with NativeAccent Pitch than with classical methods. If the major long term target application of this technology to enable effective English prosody instruction succeeds, the firm will have advanced the state-of-the-art in language teaching and will have provided an important capability to a number of people especially in business contexts for whom the ability to speak more than one language is becoming increasingly important The system could be used across diverse educational settings and could be further extended to other languages
