| Abstract |
Few years ago, sentiment and emotion analysis started to have attentions in different application areas, including political analytics customer services, healthcare, gaming and human computer interaction. Aiming at contributing to this area, we propose a deep learning model enabling the accurate emotion analysis of human speech voice. Our model consists of a sophisticated convolutional neural network (CNN) that is combined with RELU and pooling layers, and Fully connected layers with Softmax layer to achieve accurate sentiment and emotion analysis of speech. The proposed model was successfully tested using 40 speech recorded files containing 140,000 words, collected from different events. The model was able to correctly classify the test speech achieved an average accuracy of 90.7% when distinguishing between three emotions (Happiness, sadness, and Neural) in speech. Such model can be leveraged for the automatic analysis of speaker engagement level in events. © 2022 University of Split, FESB. |