Emotional speech recognition (ESR) systems are essential components in modern human–computer interaction, as they enhance communication by recognizing and responding to users’ emotions. Our proposed model leverages deep learning, specifically a convolutional neural network (CNN) architecture, which is trained to identify emotions from speech data. The model extracts and combines Mel-Frequency Cepstral Coefficients (MFCC), Root Mean Square (RMS), and Zero-Crossing Rate (ZCR) features to capture a comprehensive representation of audio characteristics. We implemented a robust data preprocessing pipeline, including feature normalization and concatenation, to optimize the input for the CNN. Our results demonstrate that this approach achieved a classification accuracy of 99.31% on the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) dataset, outperforming traditional single-feature models. The findings indicate that the integration of multifeatured audio inputs significantly enhances the accuracy and reliability of emotion recognition in speech.