Feature extraction of emotions in speech using non-negative matrix factorization

Abstract

Recognition of emotion is an important part of communication. Also, emotional speech recognition is important for the efficient human-computer interactions. For those reasons, the importance of automatic emotional speech recognition has been emphasized and the research on this area has been increased in these days. In this study, we use AIBO feature set with non-negative matrix factorization (NMF) and support vector machine (SVM) as a classifier. With NMF, we make efficient feature set to recognize emotions. Moreover, though the speech expression of emotion is affected by language and culture area of the speaker, until now, most of researches on this area have concentrated on feature extraction and emotion recognition for one language or database. We examine the characteristics of emotional speech according to the speaker’s language and culture area. By this means, we expect the improvement of emotion recognition rate in speech, and compare the characteristics of emotional expressions in several languages and cultures.