Speech Emotion Recognition Based on Principal Component Analysis and Back Propagation Neural Network

Speech signal carries rich emotional information except semantic information. Five common emotions, namely happiness, anger, boredom, fear and sadness,were discussed and recognized through a proposed framework which combines Principal Component Analysis and Back Propagation neutral network. The candidate parameters were refined from 43 to 11 via PCA to stand for a certain emotional type. Two neural network models, One Class One Network and All Class One Network, were employed and compared. The promising result, ranging from 52%-62%, suggests that the framework is feasible to be used for recognizing emotions in spoken utterance.