Audio segmentation based approach for improved emotion recognition

Recognition of emotion expressed by a customer in spoken speech is gaining importance because of the expansion of services industry and the need to identify the sentiment of the customer which are not available through a text channel. In this paper, we propose a simple yet effective method to enhance the recognition accuracy of emotion in spoken speech. The approach is based on segmenting the audio into smaller, fixed length, segments and recognizing the emotion of each segment individually before fusing the identified emotions to mark the emotion of the audio file. We show through experiments the enhanced performance of the our method on a freely available emotion database called Emo-DB.