Tree Mapping Template for Prosodic Phrase Bound-ary Predication

This paper presents a novel method driven by tree mapping template (TMT) which improve the accuracy of prosodic phrase boundary prediction. The TMT is capable of capturing the isomorphic relation between non-terminal nodes in hierarchical prosodic tree and nodes in binary tree approximation, performing pruning at the decoding phase and revising the baseline maximum entropy model with boosting method (AdaBoost). The model is statistical driven because TMTs are extracted automatically from hierarchical prosodic tree and binary tree approximation generated by the Maximum Entropy model. In decoding, TMT is employed to perform pruning, readjusting and local combination. To alleviate data sparse in limited labeled corpus, language model interpolation is made by introducing a large scale unlabeled data. The experiments show that the TMT driven method reduces the decoding complexity for the prosodic phrasing prediction which is crucial for a real-time TTS system, achieving an improvement of 11.5% in terms of F-Measure compared to a conventional maximum entropy model without TMT.