A phonetics based computer aided prosody training system for L2 English learning

The usual practice of learning L2 English prosody is a bottom up process, from initializing a word stress model followed by layering over effects from semantic and syntactic specifications, and finally paragraph association. The present study derives normalized 3-way stress models of F0 and duration from speech data with the above specifications of both L1 and L2 English to see where the differences are and how deviations from the norm that may inhibit intelligibility could be improved. The results are a word stress model achieved by teasing apart the contributions form sentential/paragraph effects. The normalized results reveal more distinct L1-L2 differences both in word phonology (categorical contrasts) and in larger speech units (sentence and paragraph); and how a model with these results could help L2 learning of English prosody.