A device, computer program and related method for processing a first seismic signal that includes identifying one portion of a second seismic signal and determining a length of a seismic wavelet. It is also possible to train a neural network by using a plurality of sub-portions of said portion a input variables and at least one second piece of information as a target variable. Said sub-portions of the portion have a length dependent on the length of the seismic wavelet determined. Finally, the method includes determining at least one first piece of geological information based on the first seismic signal using said trained neural network.