Voice Quality Features for Speech Emotion Recognition

This paper investigates the impact of voice quality features on speech emotion recognition using different sets of voice quality features alone and in combination with prosodic and spectral features. The experiment in this work involved a total of fifteen sets of emotion features, comprising three sets of voice quality features, one set of prosodic features, one set of spectral features and ten sets of hybrid features. The experimental data used in the work were taken from the Berlin Database of Emotional Speech. The classification of emotional speech was performed by using the multi-layer perceptron neural network and support vector machine techniques.