Speech-to-face movement synthesis based on HMMS

The paper describes a talking face generation system with natural and communicative reality. If face movements are synthesized well enough for natural communication, a lot of benefits will be brought for the human-machine communication. We have already proposed a speech driven HMM-based lip movement synthesis, and also shown that the quality is drastically improved by considering succeeding visemes. The paper describes extension of our system to full face movement generation, and proposes a method considering both of preceding and succeeding viseme contexts. The experiments show the proposed method generates natural and accurate talking faces from audio speech inputs.