Date of Award

Summer 8-2016

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computing

School

Computing Sciences and Computer Engineering

Committee Chair

Chaoyang Zhang

Committee Chair Department

Computing

Committee Member 2

Shaoen Wu

Committee Member 2 Department

Computing

Committee Member 3

Nan Wang

Committee Member 3 Department

Computing

Committee Member 4

Zhaoxian Zhou

Committee Member 4 Department

Computing

Committee Member 5

Huiqing Zhu

Committee Member 5 Department

Computing

Abstract

Emotion plays an important role in human beings’ daily lives. Understanding emotions and recognizing how to react to others’ feelings are fundamental to engaging in successful social interactions. Currently, emotion recognition is not only significant in human beings’ daily lives, but also a hot topic in academic research, as new techniques such as emotion recognition from speech context inspires us as to how emotions are related to the content we are uttering.

The demand and importance of emotion recognition have highly increased in many applications in recent years, such as video games, human computer interaction, cognitive computing, and affective computing. Emotion recognition can be done from many sources including text, speech, hand, and body gesture as well as facial expression. Presently, most of the emotion recognition methods only use one of these sources. The emotion of human beings changes every second and using a single way to process the emotion recognition may not reflect the emotion correctly. This research is motivated by the desire to understand and evaluate human beings’ emotion from multiple ways such as speech and facial expressions.

In this dissertation, multi-sensory emotion recognition has been exploited. The proposed framework can recognize emotion from speech, facial expression, and both of them. There are three important parts in the design of the system: the facial emotion recognizer, the speech emotion recognizer, and the information fusion. The information fusion part uses the results from the speech emotion recognition and facial emotion recognition. Then, a novel weighted method is used to integrate the results, and a final decision of the emotion is given after the fusion.

The experiments show that with the weighted fusion methods, the accuracy can be improved to an average of 3.66% compared to fusion without adding weight. The improvement of the recognition rate can reach 18.27% and 5.66% compared to the speech emotion recognition and facial expression recognition, respectively. By improving the emotion recognition accuracy, the proposed multi-sensory emotion recognition system can help to improve the naturalness of human computer interaction.

Share

COinS