Human Emotion Detection from Audio Using Feed-Forward Neural Networks

In this manuscript, we propose a novel approach for human emotion detection using feed-forward neural networks. Initially, it involves extracting Mel Frequency Cepstral Coefficients (MFCC) features from speech recordings, followed by preprocessing steps including normalization, feature extraction, and feature scaling. The dataset is then split into training and testing sets, and both LSTM and neural networks are trained to classify emotional states in audio. Through evaluation, we achieve high classification accuracy, demonstrating the effectiveness of our approach. Consequently, it holds potential applications in mental health diagnosis, enhancing customer service experiences, speech therapy, and developing more empathetic human- machine interfaces. The proposed method enhances the advancement of automated systems for human emotion recognition, opening new possibilities for improving human-machine interactions and emotional understanding.