Seismic data are often missing the lowest frequencies of sound waves that are essential for imaging deep underground structures. Capturing these low frequencies requires special equipment and can be expensive. In this work, we trained a machine learning model to learn the relationship between high and low frequencies using only simulated (synthetic) data. We then applied the trained model to real seismic data recorded using both standard (airgun) and special low-frequency (TPS) sources. Even though the model was trained only on synthetic examples, it was able to accurately predict the missing low-frequency parts of the airgun data. These predictions closely matched the actual low-frequency signals recorded by the TPS, which serves as the ground truth. This result shows that we can potentially enhance older or low-cost datasets by adding back critical low-frequency information using machine learning. The method is fast, requires no additional equipment, and could be used to improve seismic imaging for energy exploration, water resource studies, or earthquake research. By restoring missing low-frequency content, this tool helps make seismic analysis more robust and accessible,especially in areas where acquiring high-quality low-frequency data is challenging or cost-prohibitive