Emotion detection has potential to revolutionize human computer interaction by interpreting emotional nuances in spoken language. The aim of this research is to reveal the nuances of human expression by carefully examining voice patterns and extracting emotions from speech datasets. Exploratory data analysis first revealed details about the features of the dataset. The preprocessing that came next included resampling to guarantee consistent sampling rates and windowing audio signals for segmentation. Key features like log-mel spectrograms, mel-frequency cepstral coefficients (mfcc), and chroma features were extracted, shaping a comprehensive hybrid feature map in a 4-D tensor format. Several deep learning models were assessed in the study, including Xception Network (XceptionNet), Residual Network (ResNet), Mobile Network V3 (MobileNetV3), and VGG16. We measured accuracy, precision, and recall metric to assess each model’s performance through rigorous experimentation. Notably, the MobileNetV3 small model excelled, achieving an impressive 88.25% accuracy in emotion detection, showcasing its efficacy.