Abstract
Speech emotion recognition is a pivotal aspect of human-computer interaction and affective computing. This research delves into the application of the Multi-Layer Perceptron (MLP) Classifier for emotion recognition within speech, focusing on the widely used CREMA-D dataset. Our investigation covers meticulous pre-processing methods, feature extraction techniques, and model configuration. Raw audio data undergoes preprocessing, with acoustic features like Mel-frequency cepstral coefficients (MFCCs) and pitch-related parameters extracted. The MLP Classifier, a feedforward neural network, is employed to categorize emotions in speech samples. Our results reveal competitive performance in recognizing a spectrum of emotions, including happiness, sadness, anger, and neutrality within the CREMA-D dataset. Our study examines the influence of di-verse hyperparameters, training strategies, and model architectures on emotion recognition accuracy. Cross-validation ensures the reliability of our findings, emphasizing the relevance of dataset selection and the potential for advancing speech emotion recognition research. This research aids in enhancing human-computer interaction systems, sentiment analysis, and emotional well-being monitoring applications, while also contributing to the discourse on optimizing this technology.