The study of Tibetan prosodic structure prediction model

Prosodic structure prediction plays a crucial role on the prosodic annotation of speech synthesis corpus as well as on improving the naturalness of synthesized speech. The paper studies Tibetan prosodic structure with Tibetan speech characteristics. Having analyzed a variety of variables that have an impact on Tibetan prosodic boundary, we obtain syllable boundary grammatical information, prosodic environmental information and other parameters which affecting prosodic boundary types as the model prediction variables. The decision tree algorithm is used to establish prosodic structure prediction model that can predict prosodic words and prosodic phrase such two prosodic structures. Its input parameters are predicting variables while the output of the target variable is for the prosodic boundary type. The model is evaluated by news annotation corpus, experiment results show good results: in the training set of 3000 sentence, the accuracy rate of prosodic word is 89.57%, the recall rate of that is 91.22%; accuracy rate of prosodic phrase is 87.20%, the recall rate is 88.92 %.