Statistical Evaluation of Speech Features for Emotion Recognition

This paper presents an emotion recognition framework based on sound processing could significantly improve human computer interaction. One hundred thirty three (133) speech features obtained from sound processing of acting speech were tested in order to create a feature set sufficient to discriminate between seven emotions. Following statistical analysis in order to assess the significance of each speech feature, artificial neural networks were trained to classify emotions on the basis of a 35-input vector, which provide information about the prosody of the speaker over the entire sentence. Extra emphasis was given to assess the proposed 35-input vector in a speaker independent framework since test instances belong to different speakers from the training set. Several experiments were performed and the results are presented analytically. Considering the inherently difficulty of the problem, the proposed feature vector achieved promising results (51%) for speaker independent recognition in the seven emotion classes of Berlin Database.