Speech Emotion Recognition (SER) facilitates the automatic identification of emotional states from speech, with applications in mental health assessment, intelligent virtual assistants, and affective computing. A significant challenge in SER is the domain shift problem, where models trained on a specific dataset exhibit poor generalization to different acoustic environments. This paper presents a supervised domain adaptation framework that integrates Mel-Frequency Cepstral Coefficients (MFCC) for feature extraction, Correlation Alignment (CORAL) for distribution alignment, and EfficientNet-B0 for emotion classification. To support supervised adaptation, CORAL is modified using a categorical cross-entropy loss function. The proposed approach is evaluated on three datasets–RAVDESS, TESS, and a custom real-time dataset comprising 100 speakers–demonstrating substantial improvements in classification accuracy and robustness under varying acoustic conditions. Experimental results confirm that the proposed method outperforms conventional deep learning models, contributing to the development of domain-independent SER systems suitable for deployment in real-world scenarios.