Emotion Recognition on Speech using Hybrid Model CNN and BI-LSTM Techniques

Abstract Speech emotion recognition is critical for many applications such as human-computer interactions and psychological analysis. Due to the inability of conventional models to capture the subtle nuance of emotional speech variations, the identification process is less effective. The development of a new hybrid model in this study presents a solution to address this problem through combining the Convolutional Neural Networks and Bidirectional Long Short-Term Memory. The combination of feature extraction and temporal context abilities is a unique value for the model. The study model led to outstanding performance reached 98.48% accuracy, 97.25% precision, 98.29% recall, and an F1-Score of 97.39%. The latter performance surpassed those of other models such as PNN model 95.56%, LSTM model 97.1%, 1-D DCNN model 93.31%, GMM model 74.33%, and Deep Learning Transfer Models 86.54%. The developed hybrid model can accurately detect and classify emotions and speech and can effectively work in real applications.