Classifier fusion for emotion recognition from speech

The goal of this work is to investigate the performance of emotion recognition using the three features of RASTA-PLP, loudness, and long term modulation features. Single classifiers utilizing only one and combinations of all three feature types are examined. The standard Berlin database of emotional speech is used to evaluate the performance of the proposed features, comprising recordings of seven different emotions. The performance is compared with earlier work. The results reveal that, using simple fusion techniques the performance could be improved significantly, outperforming other large sets of features.