Emotion Recognition from Speech Signals using Fractal Features

In early research the basic acoustic features were the primary choices for emotion recognition from speech. Most of the feature vectors were composed with the simple extracted pitch-related, intensity related, and duration related attributes, such as maximum, minimum, median, range and variability values. However, researchers are still debating what features influence the recognition of emotion in speech. In this paper, we propose a new method to recognize the emotion from speech signals using fractal dimension features. The fractal feature indicates the non-linearity and self-similarity of a speech signal. For classification and recognition purposes we used the Support Vector Machine technique. In our experiment, a standard database, the Berlin Emotional Speech Database is used as input to measure the effectiveness of our method. By using these features, the obtained results indicated our approach has provided a recognition rate approximate 77%.