improving speech emotion recognition via gender classification

Abstract

Speech emotion recognition is a relatively new field of research that could plays an important role in man-machine interaction. In this paper we use from two new spectral features for the automatic recognition of human affective information from speech. These features are extracted from the spectrogram of speech signal by image processing techniques. Also we study the effects of gender information on speech emotion recognition. Hierarchical SVM base classifiers are designed to classify speech signals according to their emotional states. Classifiers are optimized by the Fisher Discriminant Ratio (FDR) to classify the most separable classes at the upper nodes, which can reduce the classification error. The proposed algorithm tested on the well known Berlin database for the male and female speakers separately and in combination. The overall recognition rate of 43.4% is obtained for the coeducational speakers. The results show the 39.46% improvement when the gender information is used.

Keywords

• emotion recognition
• speech processing
• emotion in males and females
• spectral patterns
• harmonic energy features