Performance Analysis of CNN-Based Speech Emotion Recognition System

This paper presents an innovative approach to Speech Emotion Recognition (SER) using Convolutional Neural Networks (CNNs), aiming to enhance human-computer interaction by enabling machines to recognize and respond to human emotions accurately. The study utilizes the Toronto Emotional Speech Set (TESS), a comprehensive dataset comprising approximately 2,800 utterances with diverse emotional states. The methodology involves pre-processing speech signals to extract Mel-Frequency Cepstral Coefficients, which are then converted into spectrograms and fed into a CNN model for emotion classification. The results demonstrate the model's robust performance in recognizing emotions such as anger, happiness, fear, and sadness with high precision and accuracy. The paper discusses the impact of various hyperparameters on model performance, highlighting the trade-offs between training time and accuracy. The findings underscore the potential of CNN-based approaches in SER, offering valuable insights for future research and applications in affective computing.