Using dialogue to detect emotion segments in movies

Movie dialogue represents participants' emotion which directly affects audiences' affective reaction. This paper presents an emotion segments detection method by using dialogue emotion detection. The proposed method has three modules. Firstly, dialogue of three categories, which are plain dialogue (PD), dialogue with music (DwM) and dialogue with other sounds (DwO) are detected by Support Vector Machine (SVM) learning. Secondly, DwM and DwO are further categorized into three emotion levels, which are high, normal and low, by Hidden Markov Model (HMM) learning. Finally, those dialogues are used to locate movie emotion segments. By this method, over 85% emotion segments are detected.