If I compare the velocity of ripped DPCM samples from a NSF file exported from Famitracker with the source NSF of the DPCM samples, the ones in my exported NSF sound a lot more quiet than those contained in the original. What could cause this problem and what can be done against it?