Emotion recognition in a chatting robot

Language plays an important role in human and household service robot communication, and speech not only contains semantic information, but also involves the tone and emotion information of speakers. We have developed a chatting robot prototype which can use speech to communicate with users. The semantic information are used to produce response by this robot. However the emotion information is also an important clue to understand the users. Therefore how to get emotion state correctly by the chatting robot is an important issue. In this paper, speech emotion recognition of a specific person is implemented. In the experiment, mel-scale frequency cepstral coefficients (MFCC) and Pitch are selected as the features used in the recognition. Two emotion cases are investigated, including neutral and emotional. Two types AdaBoost algorithms: Real-AdaBoost and modest AdaBoost are chosen to construct classifiers. The average recognition rate reaches to 90%.