Emotion Detection through Human Verbal Expression Using Deep Learning Techniques

Detecting emotions from speech signals is a crucial yet intricate aspect of Human-Computer Interaction (HCI). In the field of speech emotion recognition (SER) literature, different methods are applied to identify emotions in speech signals. These include techniques involving analysis of speech and its classification. Recently, Deep Learning methodologies are proposed as a substitute for conventional SER approaches. Spectrogram and Mel-Frequency Cepstral Coefficients (MFCC) are examples of speech features that preserve low-level characteristics associated with emotions. At the same time, textual information aids in capturing semantic meaning. Both types of features contribute to different aspects of emotion detection. In this study, we delved into the enhancement of model learning through Convolutional Neural Network (CNN) based architectures and dataset augmentation. This was particularly crucial because datasets can exhibit variability, sometimes requiring additional measures to improve model performance.