Respiratory sounds are best heard with stethoscope. This is what is called lung auscultation, A technique 
used by physicians to detect the abnormal respiratory sounds and thus diagnose the underlying respiratory 
ailments. Now, these abnormal respiratory sounds are the early signs of not only the presence, but the 
progression of underlying respiratory ailments like asthma, chronic bronchitis and many more. In this 
paper, we explore a different set of acoustic biomarkers that can automatically detect these abnormal 
respiratory sounds. So the proposed system combines 2 approaches to perform the automatic respiratory 
sound. Event detection. In first approach we use the white set of discrete wavelet transformation that is 
DWT based descriptor to explore the richness of handcrafted tempo spectral representation for respiratory 
sounds. In second approach a transfer learning using pre trained audio event detection AED model is used 
to dump this stream the respiratory sound event detection task now unlike the previous deep learning based 
approaches. We here use a simple DNN model to train the DWT coefficients without any data augmentation. 
Also are proposed Deep encoded features with computationally simple architecture are able to compensate 
the lack of training samples as well as the imbalance between healthy and unhealthy class, which is one of 
the major challenges. While we deal with the medical data, both of our proposed approaches are shown to be 
complementary. In detecting the abnormal respiratory sounds from bridge cycles. We believe this is a 
promising step towards a non invasive real time diagnosis of pulmonary disease using lung auscultation 
along with automation. Our proposed system is inexpensive yet reliable, thus can impact the lives of many 
people and reduce clinical intervention.