Researches on Learning based Speech-synch Visual Speech Synthesis Technology

摘要

The paper introduces a kind of learning-based speech-synch visual speech synthesis methodology. It was a kind of methods based on the sample collecting, learning, searching and rearrangement to synthesize the speech driven lip animation. On the basis of introducing the system structure briefly, the key technologies of system were discussed respectively, which include the constructing of acoustic-to-visual mapping model based on the viseme HMMs, the searching of lip sequence strategy and the locating of the feature point algorithm. The learning based approach gives a better simulation of the real world. The synthesis reality has been improved.