Emotion Recognition from Speech Signal through DWT - LPC & Convolution Neural Network

Abstract

Tremendous development in microcircuit technology and the Web of Things has fuelled growth in virtual personal assistant devices and systems like Alexa, Siri, and Google Assistant. These virtual assistant devices receive commands through speech signals and are trained to deliver necessary actions quickly and accurately. But these virtual assistant devices are fairly trained to receive speech commands however have to enhance their emotion recognition ability to semantically method request from the user. In this work, implementation of speech emotion recognition through Discrete Wavelet Transform (DWT) – LPC and convolution neural network (CNN) is tried. Speech signals obtained from the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) dataset are processed, transformed using DWT, reduced-order linear predictive coding coefficient, and convolution neural network (CNN). The convolution neural network was enforced for training, classification, and recognition of emotion. Relatively higher recognition accuracy was obtained through DWT - LPC & Convolution Neural Network as compared with different ways revealed within the literature.