Same source question
Different masters can dominate a format comparison. Confirm the underlying source before interpreting the score.
ABX Listening Test Trial Calculator
For a two-choice test, calculate the probability of getting at least this many answers correct by random guessing. This does not prove why a listener obtained the result.
Enter one preplanned test run
The calculation assumes two choices per trial, independent answers, and no decision to stop after seeing a favorable score.
9 correct from 12
75.0% of the choices were correct.
Upper-tail probability7.3%
Chance of this many or more correct answers under the stated random-guessing model.
Reference threshold not crossed
The five-percent reference point is 10 of 12 correct. It is 1 correct answer short of that count.
This comparison does not prove why the result occurred or how large an audible difference may be.
Conditions the number depends on
Level and cues controlled
A louder sample, visible label, switching noise, or operator cue can make the run answer the wrong question.
Stopping rule fixed
Choose the trial count before testing and retain every run. Selecting only a strong attempt invalidates the stated model.
How the probability is calculated
Add the observed result and every higher result
The calculator sums the binomial probability of your correct count and all larger counts when each trial has a 50 percent chance of success. The displayed five-percent threshold is the smallest correct count whose upper tail is no greater than 0.05.
Use the result as one diagnostic
A low probability says the score is unusual under this guessing model. It does not identify the audible cue, establish comparable files, measure effect size, or make the result universal.
Document the listening setup
Record the files, source master, playback chain, level-matching method, passages, trial count, and stopping rule. Use How to Test Whether Lossless Audio Matters to You to prepare the run before counting results.
Browse the complete listening tool set for decisions that are not probability questions.
Questions about interpreting an ABX score
Does a result below five percent prove an audible difference?
No. It means the result is unusual under the calculator's random-guessing model. It does not identify the cause or measure the size of a difference.
How many trials should I run?
Choose the count before testing and stop as planned. More trials can add information, but fatigue and changing conditions can make a longer session worse.
Can I compare files from different masters?
Not for a clean format comparison. A mastering difference can dominate the result, so first confirm that both files share the same source master.