Methods, systems, and computer-readable medium to perform operations including: generating a first time-frequency spectrum of a first seismic trace from an original seismic dataset; generating a second time-frequency spectrum of a second seismic trace from an enhanced seismic dataset, where the second seismic trace corresponds to the first seismic trace; calculating a difference between the first time-frequency spectrum and the second time-frequency spectrum to generate a noise estimate in the first seismic trace; characterizing the initial noise estimate as White Gaussian Noise (WGN); calculating, based on the characterization of the initial noise estimate, a third time-frequency spectrum of a refined noise estimate; constructing, based on the first time-frequency spectrum, the second time-frequency spectrum, and the third time-frequency spectrum, a time-frequency mask (TFM); and using the constructed TFM to generate a fourth time-frequency spectrum of an output trace that corresponds to the first and second seismic traces.