Automation is a core component of the Industry 4.0 concept and is gaining popularity among manufacturing firms. This is where the Internet-of- Things (IoT) takes place and machine monitoring system is born. The growth of the Internet of Things (IoT) has led to an enormous number of connected devices being used all over the world. It was noted that the IoT generates a huge amount of data that requires high network bandwidth, energy and large storage space, which are very costly. Therefore, many time-series data reduction methods have been introduced. This work aims to design a sound compression technique that can streamline the process of data acquisition and a sound recognition model that can reduce the network latency and cover all the information needed by the sound recognition model without losing too much accuracy. The proposed method reduces and compresses the sound data, achieving a compression ratio of 22.04. Then, the sound data will be transferred for sound recognition model training. The experiment shows that the model trained using the compressed data can achieve a high accuracy. Although the experimental results show high accuracy, further research and investigation are still needed to provide a better and more mature technique for the use cases. © 2023 IEEE.