Speech Emotion Recognition(SER) Using CNN, Fully Connected Layers, BiLSTM, and Bayesian Optimization for Parameter Tuning

In this paper, we propose a convolutional neural network (CNN) and Bidirectional Long Short-Term Memory (Bi-LSTM) based deep learning for speech emotion classification, which uses the MFCC to extract features. Data augmentation and the merging of popular datasets, RAVDESS (Ryerson Audio-Visual Database of Emotional Speech and Song) and TESS (Toronto Emotional Speech Set) used to enhance generalization, while Bayesian Optimization tunes hyperparameters. Experimental results of the proposed model achieved 99.77% training accuracy, 97.78% validation accuracy, and 98.34% testing accuracy, demonstrating the effectiveness of the model.