Respiratory infections pose a significant global health burden, with cough being a prominent symptom across various illnesses. Our paper aims to revolutionize respiratory disease screening using an automated system powered by Convolutional Neural Networks (CNNs). The system screens for bronchitis, asthma, pneumonia, and pertussis by analyzing raw cough audio. Mel-Frequency Cepstral Coefficients (MFCC) spectrograms are extracted from cough data, enabling accurate symptom identification. To ensure robustness, we augmented the dataset from 291 to 1455 files through noise addition and time stretching. The CNN architecture includes convolutional and max-pooling layers for feature extraction, with dropout regularization to prevent overfitting. This approach ensures reliable performance and adaptability to real-world conditions. With a precision of 85.2% and test accuracy of 84.2%, the model shows great potential for integration into healthcare settings, addressing the global burden of respiratory illnesses. Future enhancements could allow broader accessibility, especially in remote or underserved areas.