Cardiovascular disease continues to be a prominent cause of death globally, highlighting the need for precise and effective prediction techniques to identify and address it at an early stage. This paper introduces a new method for predicting cardiac disease by producing audio signals based on heartbeats and using ResNet-50, a deep learning framework, for classification. The performance of this strategy is compared to an ensemble decision tree algorithm. Given the fact that more and more wearables may potentially record sounds including heart sounds, audio signals can be viewed as a source of information for model building. It involves the preprocessing of unprocessed heart sound data to determine some important attributes and feed them to a ResNet-50 model for case appropriate diagnosis of various instances of heart diseases. Furthermore, we use the ensemble decision tree method to compare the results. This way, performing experiments with actual samples of heart sound, we demonstrate the efficiency of the proposed technique in predicting illness in the hearts. Furthermore, we evaluate the performance of ResNet-50 to ascertain how well it performs when compared with the ensemble decision tree approach in conditions of accuracy, sensitivity, specificity and computing efficiency. In the light of the present study, it may be concluded that the proposed approach which is developed based on the ResNet-50 model outperforms the ensemble decision tree algorithm in the aspect of accuracy. This shows that our approach can be a reliable technique used in estimating heart disease. To sum up, the present work contributes to the development of heart disease prediction by up to date technology and methodologies and optimizes the model for efficient and accurate prediction.