Numerous IoT systems generate substantial and diverse data, necessitating rapid and on-demand analysis. A primary concern revolves around the substantial energy consumption resulting from data transfers to the cloud. Leveraging machine learning algorithms, the distribution of workloads between the cloud and closer data sources is achieved, optimizing processing efficiency. This approach not only saves time but also enhances privacy and reduces network traffic. This study proposes an energy-efficient approach for gathering and analyzing IoT data. Prior to transmission, this research employs a lossless compression based on Huffman adaptive encoding on the deployed data to conserve the energy of the IoT devices. Subsequently, the transmitted data is reconstructed and processed using supervised deep-learning methodologies. The test results illustrate a reduction in transmitted data by multiple factors, showcasing significant data savings without compromising the accuracy or quality of the audio data.