Improving prosodic phrase prediction by unsupervised adaptation and syntactic features extraction

In the state-of-the-art speech synthesis system, prosodic phrase prediction is the most serious problem which leads to about 40% of text analysis errors. Two optimization strategies are proposed in this paper to deal with two major types of prosodic phrase prediction errors. First, unsupervised adaptation method is proposed to alleviate the mismatching problem between training and testing. Second, syntactic features are extracted from parser and integrated into prediction model to ensure that the consistency between the predicted prosodic structure and the syntactic structure. We examine our methods on an in-house Mandarin speech synthesis system and experiment results show that both strategies yield positive effects and the sentence unacceptable rate significantly drops from 15.9% to 8.75%.