Chinese Prosodic Word Prediction Using the Conditional Random Fields

Abstract

As the basic prosodic unit, the prosodic word play an important role for the naturalness and the intelligibility for the Chinese TTS system. Although many research work have been on this research direction, the precision of the prosodic word prediction is still not satisfying. In this paper, Conditional random fields is introduced to model the prosodic predicting process. In this model, more efficient features can be fused together. Compared with the ME model, the CRF model can describe the interacting relations between the neighboring prosodic words. The experiment results show that this conditional random fields model is competent for the prosodic word prediction task. The f-score of the prosodic word boundary prediction reaches 96.81%.