Emotion Recognition from Speech Using Perceptual Features and Convolutional Neural Networks

Abstract

Emotional computing has played a crucial role in acting as an interface between humans and machines. Speech based emotion recognition system is difficult to be implemented because of the dataset which is containing a limited number of speeches. In this work, multi speaker independent emotion recognition encompasses the use of perceptual features with filters spaced in BARK scale and Equivalent rectangular bandwidth (ERB) and vector quantization (VQ) for classifying groups and convolutional neural network with backpropagation algorithm for emotion classification in a group. The proposed system has provided consistently better accuracy for the perceptual feature with critical band analysis done in ERB scale with overall accuracy as 86% and decision level fusion classification yielded 100% accuracy for all emotions. Speaker dependent emotion recognition system has provided 100% as accuracy for all the emotions for ERB-PLPC features and perceptual linear predictive cepstrum has given 100% as accuracy for all emotions except sad emotion.