<![CDATA[Traditional methods for evaluating lecturer performance in education, such as student surveys, are often limited by their nature. This study explores the development of an objective, a framework to complement these evaluations through Speech Emotion Recognition (SER). This Research utilizes a specialized Indonesian speech emotion dataset, applying data augmentation techniques to enhance model generalization. A set of advanced acoustic features, including Mel Frequency Cepstral Coefficients (MFCCs), Chroma, and Spectral Contrast, along with their statistical variations, is used to create representations of the vocal expressions. A Multi Layer Perceptron (MLP) neural network was designed and trained on these features to classify five different emotions: happy, angry, sad, surprised, and neutral. The Research resulted in a model that demonstrated very good performance, achieving an overall classification accuracy of 94% with high precision, recall, and F1-scores across all emotions, indicating a balanced and reliable system. A critical feature analysis was also conducted, revealing the significance of the standard deviation of Chroma and MFCC features. This study shows that an MLP model paired with feature engineering can be used as a powerful and objective tool for providing deeper insights into student feedback, contributing a valuable new methodology for quality assurance in higher education.]]>
Copyrights © 2025
