Automatic Prosody Boundary Labeling of Mandarin Using Both Text and Acoustic Information

Prosody is an important factor for a high quality text-to- speech (TTS) system. Prosody is often described with a hierarchical structure. So the generation of the hierarchical prosody structure is very important both in the corpus building and the real-time text analysis, but the prosody labeling procedure is laborious and time consuming. In this paper, an automatic prosody boundary label system is presented, in which the classification and regression tree (CART) framework is used. In this system, we build a prosody model using acoustic information and the text information based on large speech corpus with prosodic structure label (ASCCD). Experiments show this model can achieve prosody boundary detection 90.86% accuracy.