COVID-19 pandemic outbreak has taken the world by storm in the 18 months and the ramifications are by no means curtailing. The need of the hour with COVID-19 and other pulmonary diseases is a quick online diagnosis by handheld devices. In the light of these constraints, scientists are relying on audio based automated techniques since clinicians routinely use audio cues from the human body (e.g. vascular murmurs, respiration, pulse, bowel sounds etc.) as markers for diagnoses of diseases or the development of ailments. Until recently, such signals have been commonly obtained during scheduled visits via manual auscultation. Research has also begun to use digital technologies to collect body sounds for cardiovascular or respiratory tests, e.g. from stethoscopes, which can then be used for automated artificial intelligence- based analysis. An early study has promised to detect COVID-19 from cough and speech diagnostic signals. This research work describes how preprocessing techniques can enhance the performance of a methodology established over a large-scale crowd-sourced dataset of respiratory audios and in what ways preprocessing techniques ameliorate the performance of cough based diagnosis. Our findings demonstrate that a machine learning classifier will better distinguish a healthy individual from individual with cough due to bronchitis, pertussis or COVID-19 by applying preprocessing techniques. Robust results have been procured by user-based data split-up for the K-fold learning methodology. The results show a noticeable increase in the efficacy of the application of preprocessing techniques in an algorithmic pipeline. These results are rudimentary and only the tip of the iceberg of the potential of cough and audio-based machine learning. The research opens the door for enhancing the performance of lightweight machine algorithms to be comparable with their more complicated and resource-consuming counterparts. Such advancements can be of paramount significance in the practical field of application deployment.