Multi-expression Facial Animation Based on Speech Emotion Recognition

A facial animation generation system is implemented based on MPEG-4 standard according to both multimedia ring and the emotion embedded. By choosing hidden Markov model as the classifier, five kinds of emotions, including anger, gratulation, cute, puzzle, excitement, are generated in speech database through training, and then a set of expression FAPs for each kind of emotion is built up. By analyzing the intensity of speech signal, synthesis expression function is obtained and used to integrate expression FAP with lip animation FAP to obtain the synthesis FAP for driving facial mesh to generate animation. The test result indicates that the maximum recognition ratio of our system may achieve 94.44%, besides its realistic facial animation.