Investigating Pitch Accent Recognition in Non-native Speech

Gina-Anne Levow

Abstract: Acquisition of prosody, in addition to vocabulary and grammar, is essential for language learners. However, it has received less attention in instruction. To enable automatic identification and feedback on learners' prosodic errors, we investigate automatic pitch accent labeling for non-native speech. We demonstrate that an acoustic-based context model can achieve accuracies over 79% on binary pitch accent recognition when trained on within-group data. Furthermore, we demonstrate that good accuracies are achieved in cross-group training, where native and near-native training data result in no significant loss of accuracy on non-native test speech. These findings illustrate the potential for automatic feedback in computer-assisted prosody learning.