Detecting Human Emotions in a Large Size of Database by Using Ensemble Classification Model

One of the most challenging researches in the field of Human-Computer Interaction (HCI) is Speech Emotion Recognition (SER). Several factors affect to the classification result. For example, the accuracy of detecting emotion depends on type of emotion and number of emotion which is classified and quality of speech is also the importance feature. Four different emotion types (anger, happy, natural, and sad) from Thai speech was used in this research. All of theses speech were recorded from Thai drama show which were most similar with daily life speech. The ensemble classification method with majority weight voting was used. This proposed algorithms used the combination of Support Vector Machine, Neural Network and k-Nearest Neighbors for emotion classification. The experimental results show that emotion classification by using the ensemble classification method by using the majority weight voting can efficiency give the better accuracy results than the single model. The proposed method has better results when using with fundamental frequency (F0) and Mel-Frequency Cepstral Coefficients (MFCC) of speech which give the accuracy results at 70.69.