This study presents a comprehensive framework for the automated analysis of emotional tone in video lectures, combining modern audio-processing libraries, feature-extraction techniques, and supervised classification methods. We evaluated open-source tools—Librosa for spectral and cepstral analysis, Praat for phonetic parameters, and PyDub for flexible audio manipulation—to extract over 100 acoustic features, including Mel-Frequency Cepstral Coefficients (MFCCs), chroma statistics, signal-strength metrics (maximum, mean, and average loudness), and prosodic cues such as pitch and intensity. Unsupervised clustering with K-Means on both spectral and signal-strength feature subsets achieved initial cluster accuracies of up to 90% for non-emotional and 79% for emotional segments. Building on these results, we trained a Support Vector Machine (SVM) with an Radial Basis Function (RBF) kernel on selected feature subsets, obtaining a baseline classification accuracy of 83% on unseen data, which improved to 89% upon inclusion of additional MFCC coefficients. To streamline the labor-intensive annotation process, we developed a web-based tool using Wavesurfer with customizable hotkeys, reducing manual tagging time. Our findings demonstrate that combining spectral, chromatic, and signal-strength features with SVM classification provides a robust approach for detecting emotional fragments in lectures. These results have important implications for enhancing pedagogical feedback systems and student engagement, and point toward future extensions involving deep-learning models and real-time analysis.