“Epochs” are equal-duration chunks of the continuous raw signal. Epochs are created relative to a series of “events” (an event is a sample number plus an event ID integer encoding what kind of event it was). “tmin” and “tmax” define the start and end times of the epochs, relative to the times of the events. So in the example you quoted, for each event, an epoch is created around it that starts 1.5 seconds before the event and ends 4.5 seconds after the event. Later, the extra 0.5 seconds are cropped off the beginning and end after computing the time-frequency representation of the epochs. The extra 0.5 seconds was originally included so that when computing the time-frequency representation, edge effects would not occur during the times of interest.