Speech emotion recognition using cross correlational database with feature fusion methodology

Abstract

Speech emotion recognition provides an interface for communication between the human and the machine. Classifying the emotion based on the speech signals is not that easy task since we need to take into account the conditions like noisy data, changes in voice due to cold/cough and so on because the voice of a person will not be the same when he/she is suffering from cold/cough, or when he/she consumed alcohol. In this paper we just extracted some of the features like Volume, Energy, MFCC, and Pitch in order to classify the emotion into happy/sad/anger/neutral. In this paper MFCC plays a major role for classifying the emotions into happy/anger/sad/neutral. The concept of Cross-Correlation is that we first make use of Berlin Database and train the model using Berlin database and then we will test the same model using Spanish Database. The main role is that to test whether the model taken produces the same output (emotion) for both the Spanish and Berlin Databases that is we need to prove that the model taken is independent of the language used. Accordingly, a function is developed in MATLAB for Identification of an Emotion for any Audio File given as an input [1].