Human Speech Based Emotion and Gender Recognition Using Hybrid SVM and Deep Learning Technique

This study uses the RAVDESS Dataset to develop a hybrid model for predicting gender and emotion from human voice. To improve emotion categorization accuracy, the suggested system uses a two-stage methodology. Our goal is to predict the emotion as well as gender with higher accuracy in two different stages as mentioned below. A SVM model is used in the initial step of gender prediction to determine if the speaker is male or female. In the second step, a convolutional neural network (CNN) included with a recurrent neural network (RNN) model is used to forecast the speaker's mood, and the output of the SVM model is then integrated as an extra feature. The CNN-RNN model in this process is trained with Mel spectrograms and MFCC characteristics, and studies demonstrate how well this two-step approach works to enhance emotion identification capabilities. The method identifies eight different emotions. This two-way approach helps to identify the gender and the emotion of the voice with a better accuracy