Analysis of Speech Emotion Recognition Using Deep Learning Algorithm

Abstract

In this project, we propose an automated system for Speech emotion recognition using convolution neural network (CNN). The system uses a 5 layer CNN model, which is trained and tested on over 7000 speech samples. The data used is.wav files of speech samples. Data required for the anlysis is gathered from RAVDESS dataset which consists of samples of speech and songs from both male and female actors. The different models of CNN were trained and tested on RAVDESS dataset until we got the required accuracy. The algorithm then classifies the given input audio file of.wav format into a range of emotions. The performance is evaluated by the accuracy of the code and also the validation accuracy. The algorithm must have minimum loss as well. The data consists of 24 actors singing and speaking in different emotions and with different intensity. The experimental results gives an accuracy of about 99.8% and a validation accuracy of 93.33% on applying the five layer model to the dataset. We get an model accuracy of 92.65%. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.