Enhancing Speech Emotion Recognition Through Multimodal Fusion and Advanced Feature Extraction Techniques

Accurate emotion recognition is crucial for enabling computers and intelligent systems to respond emotionally to stimuli. In this study, we implement emotion recognition by leveraging convolutional neural networks (CNNs) within a deep learning framework. Specifically, we adapt an image processing CNN architecture, implemented in Python using the Keras model-level library with the TensorFlow backend. We present a condensed discussion of the theoretical foundation underlying the classification of emotions based on voice parameters. Our findings reveal that our model achieves an average accuracy of 68% in identifying eight distinct emotions: happiness, fear, sadness, neutrality, anger, surprise, and disgust. This performance aligns with the stateof-the-art reported in the scientific literature. Notable contributions of this study include the adaptation of a deep learning model for audio file processing, CNN training using English language recordings, and the development of an experimental software environment for generating test files.