Emotion recognition is required for human behaviour and mental state study, applied in areas such as healthcare, education, and human-computer interaction. In this paper, an Artificial Intelligence-based Emotion Recognition System is proposed which employs speech analysis for emotion recognition. Unlike other approaches, our system integrates an optimized Support Vector Machine (SVM) model with hyperparameter optimization, and it achieves high accuracy at 89% on the RAVDESS dataset. The novelty of this work lies in its hybrid approach combining CNN for feature extraction and SVM for classification, enabled through Mel Frequency Cepstral Coefficients (MFCC) feature extraction and a sophisticated SVM classification method. This enables real-time stress measurements and emotional surveillance of students, supporting mental wellness tracking in real-world environments. Additionally, our design enhances deployment feasibility across a variety of embedded platforms. The future work will explore multimodal fusion, integrating multimodal data and physiological signals to improve emotion classification accuracy. The proposed system is of very much interest for intelligent emotion-sensitive applications in medical therapy, human-machine interface, and mental illness monitoring. Additionally, datasets like IEMOCAP and CREMA-D provide more spontaneous and culturally diverse samples that can be utilized in future implementations.