This study aims to provide a comprehensive analysis of the distribution of respiratory sounds using a large number of recordings. The data set that we used includes normal breathing sounds, wheezing sounds, and coughing sounds, all of which indicate a different breathing pattern. The method involved first removing audio data and making the data consistent, then extracting features using Mel Frequency Cepstral Coefficients (MFCC), Based on these features, breath sounds are classified using a convolutional neural networks (CNN) model. The CNN model has shown high accuracy in distinguishing all the various respiratory sounds, showing and evaluating that it could be useful in diagnosis. This work has demonstrated how well CNN can identify breath sounds and suggests ways to develop automatic, non-invasive respiratory monitoring tools to help in the field of respiratory problems.