Speech emotion detection based on neural networks

Emotion detection in spoken dialogues is an area that has traditionally been studied in psychology and linguistics but in recent years the engineering community has become increasingly active in this area, due largely to its importance in spoken language man-machine interfaces. Besides techniques in signal processing and analysis it also requires psychological and linguistic analysis. This paper reports an experimental study on six emotions, happiness, sadness, anger, fear, neutral and boredom. It uses speech fundamental frequency, formants, energy and voicing rate as extracted features. Features are selected manually for different experiments in order to get the best results. The selected features are included into a features vector with different sizes as input for different neural network classifiers. To carry out this experimental study a specific tool for language-independent emotion recognition tool has been designed and used. The database which is used for this experiment is the Berlin Database of Emotional Speech.