A huge development is done in current living in the disciplines of artificial intelligence, machine learning, human-machine interaction, etc. It is becoming more and more usual to interact with machines or give them instructions for specific tasks using voice commands. Siri, Alexa, Cortana, Google Assist, etc. are embedded into a lot of consumer electronics. But machines are constrained because they can't communicate with people the same way that people can. It cannot comprehend or respond to human emotions. Research in recognising emotions from speech is being led by the field of human-machine interaction. A more robust man-machine communication system is necessary given the significance of machines in our everyday lives. The goal of speech emotion recognition (SER), which is being developed by several researchers, is to improve communication between humans and machines. To do this, a machine must be able to identify emotional states and respond to them in a manner similar to how we humans do. The calibre of the retrieved features and the kind of classifiers employed determine how efficient the SER system is. The four main emotions—anger, sadness, neutrality, and happiness—from speech were the focus of this investigation. In this study, distinct emotions are identified using convolutional neural networks (CNNs) and the Mel Frequency Cepstral Coefficient (MFCC) approach for extracting characteristics from voice. Finally, the simulations showed that the proposed MFCC-CNN outperformed existing models in terms of performance.