A Comparison of Neural Networks for Real-Time Emotion Recognition from Speech Signals

Speech and emotion recognition improve the quality of human computer interaction and allow easier to use interfaces for every level of user in software applications. In this study, we have developed two different neural networks called emotion recognition neural network (ERNN) and Gram-Charlier emotion recognition neural network (GERNN) to classify the voice signals for emotion recognition. The ERNN has 128 input nodes, 20 hidden neurons, and three summing output nodes. A set of 97920 training sets is used to train the ERNN. A new set of 24480 testing sets is utilized to test the ERNN performance. The samples tested for voice recognition are acquired from the movies “Anger Management” and “Pick of Destiny”. ERNN achieves an average recognition performance of 100%. This high level of recognition suggests that the ERNN is a promising method for emotion recognition in computer applications. Furthermore, the GERNN has four input nodes, 20 hidden neurons, and three output nodes. The GERNN achieves an average recognition performance of 33%. This shows us that we cannot use Gram-Charlier coefficients to discriminate emotion signals. In addition, Hinton diagrams were utilized to display the optimality of ERNN weights. Key-Words: Back propagation learning algorithm, Neural network, Emotion, Speech, Power Spectrum, FastFourier Transform (FFT), Bayes optimal decision rule.