Speech-Based Emotion Recognition Using Acoustic Features and Neural Networks

Speech Emotion Recognition (SER) enables systems to identify emotions from spoken language by analyzing acoustic features like pitch and intensity. This paper presents an SER system using the Toronto Emotional Speech Set (TESS) to classify seven emotions: anger, disgust, fear, neutral, happiness, pleasant surprise, and sadness. Features were extracted using Mel-Frequency Cepstral Coefficients (MFCCs) and classified using Random Forest model with precision as 1.00, recall as 0.98, 97.5% accuracy and CNN model with 99.64% accuracy. An interactive application built with Streamlit enables real-time emotion recognition, demonstrating the system's potential for enhancing human-computer interactions. The experimental results demonstrate the potential of the proposed SER system to accurately classify emotions, paving the way for more empathetic and natural interactions in technology.