Using the Fisher Vector Representation for Audio-based Emotion Recognition