For this approach, the input utterances were assumed contained more than one emotion as mentioned in previous section. The utterances were segmented into pieces of emotions according to the experiment in the next section. The system was also allowed to answer difference emotions in an utterance freely. For example, the input utterance was divided into five segments and all of segments were labeled as “Fear”, but the output of classification might independently answers five different emotions like “Fear, Sad, Sad, Fear, and Fear” as displayed in Figure 2.