Shouldn't unsigned int duration be unsinged long duration?
I replaced it in your lib. So that it can work even TCCR0B = TCCR0B & B11111000 | B00000001.
Setting clock 0 to this makes one second last 64000 instead of 1000 millis. This causes your function to overflow. If you are working with millis() than you should use unsigned long not unsigned int.