Performance Analysis of Human Emotion via Speech Recognition using Convolution Neural Network Algorithm compared with Hidden Markov Model Classifier for Improved Accuracy

This research intends to use a new Convolution Neural Network method, as opposed to the old Hidden Markov Model (HMM) algorithm, to make better predictions about people's conveying feelings by the noises they make. The dataset for this research is taken from Toronto Emotional Speech Set (TESS). Using G-power 0.8, we determined the sample size for each dataset. The prediction of human emotion identification from voice signals is performed using either a Convolution Neural Network or a Hidden Markov Model, both of which need the same number of data samples (N=10). The suggested Convolutional Neural Network performs significantly better with the accuracy rate of 93.68% than the accuracy obtained by the Hidden Markov Model Classifier, which infers Convolutional Neural Network performs better. The significance level obtained by the investigation was p = 0.001 (p<0.05) and the two groups are statistically significant. When comparing the two models' performance in human emotion categorization using voice data, the suggested CNN model outperforms the Hidden Markov Model (HMM).