Find out whether a difference is audible to you, on your system, rather than whether it exists. A null test shows what differs between two files; an ABX test asks whether your ears can tell. The two are different questions, and this page answers the second.
Preparing the pair
Load the two files as A and B. Prepare runs one pass over both: the offset between them is measured by cross-correlation and B’s level is matched to A’s. Both corrections can be turned off, for a pair that is already matched or when the level is the thing under test.
The trials
Each trial, X is one of the two, chosen before you listen. Play A, B and X as often as you like; switching keeps the same position in the music, so you compare the same notes. Answer X is A or X is B, see whether you were right, and move on. Loop start and length set the stretch of music that repeats; a few seconds of a telling passage beats a whole song.
Reading the score
The verdict is the probability of scoring at least your hits by guessing, from the binomial distribution at one half. Below 1 percent you can hear it; below 5 percent you probably can, and more trials will settle it; above that the two are not told apart under these conditions. The run can be copied as JSON with every answer.
Honest conditions
Use headphones, keep the level the same across the run, and do not look at the waveforms while answering. A difference that is obvious sighted and vanishes blind is the finding the test exists to make.