Emotion Recognition using Fuzzy Clustering Analysis
Tejal Udhan · 2016
This research project investigates using fuzzy clustering algorithms for emotion recognition. Emotion recognition has gained significant attention in recent years in applications such as artificial intelligence, human-computer interaction, speech and voice recognition. The ability of a computer or machine to understand human emotion and respond to users in a more human way can lead to significant advances in conversational speech recognition systems, improved quality of life in persons with speech disorders, such as Parkinson’s disease and even in voice response systems, such as Google Voice or Apple’s Siri. Experimental results in this area can inform discovery and innovation of machine intelligence and actionable response algorithms that use physiological methods for characterizing speech. Human emotion is a complex signal that is difficult to characterize analytically. One proposed method for characterizing emotion is to use fuzzy clustering techniques to partition the data into classifications of emotions based on feature similarities. Fuzzy clustering provides a method for organizing data into groups either in unsupervised fashion or based on the selected feature and classifying each group as a different emotion. In this work, an emotional prosody speech dataset is prepared input to a fuzzy clustering toolbox to explore underlying structures in the dataset and perform data reduction for optimal feature extraction. The emotion dataset includes fifteen different categories of emotions: happy, elation, sadness, despair, boredom, interest, shame, pride, contempt, disgust, panic, anxiety, hot anger, cold anger, and no emotion. The primary goal of this research project is to identify a fuzzy clustering technique that will partition the dataset into different categories of emotions. Background and Objectives • The interaction between human beings and computers will be more natural if computers will be able to perceive and respond to human verbal communication or non-verbal communication. Verbal mode has speech i.e. acoustic signals where as non-verbal mode contains facial expressions. • Past researches show that use of either of the modes of communication for emotion recognition is not efficient (only between 50% to 70% ). Neural networks, hidden Markov models (HMMs) and support vector machines (SVMs) are the few types of algorithms that has been used. • Since Fuzzy clustering is one of the powerful tools in data mining and has supervised and unsupervised algorithm availability, it performs well in situations where there is large variability in the data clusters. Consequently, it may should provide a good platform for emotion recognition algorithm. The main objectives of this project are two-fold: (1)To determine an acoustic feature that can characterize human emotion efficiently. (2)To develop a clustering algorithm performing emotion recognition within the dataset that has closely related emotions