Intelligent robotic systems need to be aware of humans if these humans' activities are not to be disturbed by the robot's presence. When a robot is performing an activity, the acoustic noise that it produces can be unpleasant for the people around. To avoid disturbing people, the robot may regulate its motion-induced acoustic noise by adapting its motion speed. In this paper we propose a solution to this problem that allows the robot to select, at each moment, the maximum velocity possible (up-to a reference desired velocity) that induces less acoustic noise than the environment's background acoustic noise. The motion-induced acoustic noise for each velocity-environment tuple is actively learned by the robot during a learning phase. A set of experimental results shows the ability of the proposed solution to allow an indoor wheeled robot to adjust its velocity according to the background acoustic noise.