I did a little reading, and it all comes down to something called "zero stretching" which is essentially breaking the conventional constant-voltage DC power into a stepwise series of square-wave power pulses.
In DCC, power sent to the rail is a generally-symmetric square wave, i.e., short positive pulses followed by short equal-and-opposite negative pulses. The TIME of these pulses is used to send control signals to the trains, much like the funny bar-code things that we see at the store when we buy anything.
However, with "zero stretching", an intentional "break the rules" trick is played ... the duration of a positive pulse is stretched so that it is longer than it's negative-pulse mate. They don't cancel each other, but rather add a net sum greater than zero (in the positive voltage direction). That makes a conventional DC loco go. To go faster, the DCC system extends the biased offset between the positive and negative pulses to yield an even-longer net sum greater than zero.
But all the while, the traditional DC loco is STILL seeing (and receiving) a rapid series of positive and negative pulses ... hence, the "humming noise". And hence, the rationale that this is ultimately bad for the motor to leave the loco sitting idle on the track for any duration.
TJ's disclaimer ... please don't think that I really know what I'm talking about! I'm just babbling after a read of the info below, at this link ...
Hope this helps!