COMPARISONS AMONG FOUR STATISTICS BASED METHODS OF PROSODY STRUCTURE PREDICTION
会议名称：《第七届全国人机语音通讯学术会议（NCMMSC7）》
会议日期：2003年
摘 要：Prosody structure prediction plays an important role in text-to-speech (ITS) conversion systems, ft is the must and prior step to parametric prosody prediction. Dynamic programming (DP) and decision tree (DT) are widely used for prosody structure prediction [1][2][3] but with well-known limitations. In this paper, two other new methods, combination of dynamic programming with decision tree and combination of decision tree with finite state machine (FSM), are proposed. Then, based on a manually labeled corpus, comprehensive comparisons among the four methods are done. It could be concluded from these experiments that combination of dynamic programming with decision tree method is the best choice for prosody word boundary prediction and combination of decision tree with FSM is the best candidate for prosody phrase boundary prediction.