关键词: 维吾尔语, 韵律边界, 分层预测, 独立特征模板, 形态特征

Abstract: Accurate prediction of prosodic boundary is a very important task in an arbitrary Text-To-Speech（TTS） system. A two-layer bottom-up hierarchical approach based on conditional random fields is used to predict prosodic word and prosodic phrase boundaries according to the prosodic hierarchy of Uyghur language. Morphological features are also integrated into the feature sets which are used to train the prosodic boundary prediction model. In order to further enhance the accuracy of prosodic boundary prediction, a two-layer bottom-up hierarchical approach is improved by selecting independent feature sets at every layer according to the different prosodic boundary structure of corresponding layer. Consequently, the best prosodic boundary prediction performance is achieved by extensive and repeated experiment with different feature sets and different models. Experimental results show that the proposed method obviously improves the prediction of prosodic boundary.

Key words: Uyghur language, prosodic boundary, hierarchical prediction, independent feature template, morphological feature