Emotion recognition in spontaneous Filipino speech using machine learning classification

Empathic computing features allowing computers to be able to identify the emotions of a user and give feedback according to these emotions. A lot of research effort has been dedicated to different techniques that may be used so that a computer may correctly identify human emotions. A popular approach to the problem has been through machine learning algorithms. Studies train systems to perform recognition using various combinations of acted emotion, spontaneous emotion, and modality. This study focuses on identifying discriminant voice features and testing different machine learning classification algorithms to recognize the emotions of happiness, fear, neutrality, sadness and anger in spontaneous Filipino speech using voice as the modality. The algorithms used are C4.5, k-nearest neighbor, Naive Bayes, logistic regression, support vector machine and an artificial neural network using multilayer perceptrons. Of these, C4.5, produces the best recognition rate at 73% using the features gender, energy, the third formant, maximum pitch and minimum pitch