Comparative Study on Sentiment Analysis of Human Speech using DNN and CNN

Abstract

AbstractCommunication is a way of exchanging thoughts through emotions. In this paper, we have proposed a method where human speech is converted into digital input. The digitized sound is then fed into the proposed models, and the voice of every person is classified into discrete emotional characteristics by its pitch, intensity, timbre, speech rate, and pauses. In the proposed method, we have drawn a comparative study between sentiment analysis of human speech using deep convolutional neural network (CNN) and dense deep neural networks (DNNs). In this method, multiscale area attention is applied in deep CNN as well as dense DNN to obtain emotional characteristics with wide range of granularities and therefore, the classifier can predict a wide range of emotions on a broad scale classification.KeywordsSentiment analysisAudio analysisDeep learningNeural networksEmotion detectionDeep neural network (DNN)Convolutional neural network (CNN)