A vehicle sound generation device includes a controller that generates a sound signal representing a sound and sets the localization of the sound, and left and right front speakers and left and right rear speakers that output sounds corresponding to the sound signals for which localization has been set by the controller, in which the controller sets the localization of the sounds based on the change amounts (such as the pitch rate, the torque change amount, the roll rate, and steering angular speed) per unit time of the physical quantities correlated with a driving operation by the driver of the vehicle so that the sound image of at least part (low frequency component) of frequency components of the sound is located toward the wheel where the ground load is increasing according to the driving operation.