Production counters, such as mechanical/photoelectric switch sensors, line sensors, or infrared proximity sensors, are often used as real-time productivity counters for manufacturing machines. These counters can be easily tampered with and are not designed for detecting machine fault. Since manufacturing machines with mechanical operations produce sound, capturing the sound data should provide opportunities for both production counting and detecting anomalies and hence taking proactive action to reduce production wastage. But in the manufacturing environment, often there are more than one machine in operation. These machines generate different background noises at different times, which makes gathering labeled data to train machine learning a challenging task and subsequently affects the production tracking and counting accuracy. The objective of this work is to develop a robust sound-based counter to track and count the productivity of a manufacturing machine in a real environment. Our proposed solution uses the Kolmogorov-Smirnov test to detect changes in the ambient environment. Experimental results show that our approach can achieve an accuracy of up to 92 while other conventional techniques are completely disoriented. © 2023 IEEE.