The automatic prediction of Chinese text´s prosodic structure based on tree structure

The recognition of prosodic structure is an important research aspect in the field of Text-to-Speech. It is essential to improving the naturalness of machine-synthesized speech. This paper proposes an approach to predicting and assigning prosodic structure automatically for Chinese sentences based on their tree structures. It presents the modeling of a statistical language model based on the simply handled corpus, the generation arithmetic of trees, and the method for predicting prosodic boundaries based on tree structures. The experimental results show that this tree-based method is effective, and it can obtain good prediction results.