Real-time emotion recognition is critical for enhancing human-computer interaction in fast-paced service environments, yet many existing systems rely on post-processed analysis or prioritize facial expressions over vocal cues. This study addresses this limitation by developing and validating a real-time speech emotion recognition system that exclusively utilizes prosodic acoustic features, such as pitch, energy, tempo, and pause ratios, to classify emotional states. The research employed a quantitative, developmental design, collecting unscripted speech data from 352 tertiary students to ensure naturalistic emotional expression. After rigorous preprocessing and feature selection to optimize computational efficiency, the system’s performance was evaluated across three machine learning algorithms: Random Forest, XGBoost, and Support Vector Machine (SVM). The results indicate that the Random Forest model achieved the highest predictive accuracy of 50.00%, significantly exceeding the 20% random baseline for a five-class classification problem. While this reflects moderate predictive performance, the findings highlight the inherent limitation of relying exclusively on prosodic features without spectral or semantic augmentation. Notably, the system demonstrated exceptional real-time capability, achieving an extraction rate of 20 to 35 files per second and near-instantaneous feedback delivery, thereby emphasizing its practical utility as a responsive, real-time emotion-aware support tool rather than a purely predictive model. Furthermore, evaluations based on the ISO/IEC 25010 software quality model yielded highly favorable scores from both general users (4.58) and software experts (4.59), particularly in functional suitability and usability. These findings confirm the system’s operational viability for real-world deployment, such as in Registrar’s Offices, where it can provide instant emotional insights to help service providers respond more empathetically and prevent communication escalation. While the system is technically robust, future iterations should integrate spectral elements like Mel-frequency cepstral coefficients (MFCCs) to enhance classification accuracy and generalizability across diverse populations. By providing immediate, non-invasive emotional feedback, this research contributes to the development of human-centric technology that is responsive to the psychological needs of individuals in professional settings.

Keywords: Speech Emotion Recognition (SER), prosodic features, real-time processing, machine learning, human-computer interaction, ISO/IEC 25010