In this paper, we proposed a facial expression synthesis method in speech using the Kohonen neural network (KNN) and investigate quantitatively synchronization of generated facial expression series with real speech sounds. The basis of this method is to utilize facial expression series in speech of vowel to vowel generated from the still images in speech of 5 vowels and non-speech by KNN. In synchronization with speech sound, the stabile section and the changing section of facial expression are detected by the power of speech, and a fixed facial expression image is allocated in the stabile section and facial expression series is placed with the same time interval in the changing section. In the evaluation experiment, two different synchronizing methods are compared. As this result, it is confirmed that the proposed method can represent more real face expression than the other method.
展开▼
