Can you really hear the difference between a FLAC and a 320 kbps MP3, or between two masters? Load both versions and take a proper blind ABX test: the loudness is matched, both files stay in sync when you switch, and the result comes with a statistical p-value so you know whether you heard it or guessed.
Last updated:
A
No file yet
B
No file yet
0:00.00:00
Load two files (or the demo pair) to start.
Ready for a blind test
Listen to A and B as long as you like. When you start, X is secretly A or B in each trial; switch between A, B and X at the same position and decide which one X is.
Trial
X was
You said
Result
Time
Advertisement
Advertisement
How to use the ABX Audio Test
Load the two versions as A and B, for example the FLAC and the MP3 made from it. A demo pair is loaded so you can try it right away.
Check the setup: loudness matching and alignment are automatic. Optionally set a loop around a passage with cymbals, applause or reverb tails.
Listen to A and B openly until you think you know the difference.
Press Start ABX test, then in every trial switch between A, B and X and answer X is A or X is B.
Read the result: your score, the p-value and what it means. Below p = 0.05 you very likely hear a real difference.
Scores needed for a significant result
With a fair coin, each answer has a 50% chance of being right. These are the minimum numbers of correct answers whose probability under pure guessing is below the usual thresholds (exact one-sided binomial test):
Trials
Correct for p < 0.05
Correct for p < 0.01
Example p-value
8
7 (p = 0.035)
8 (p = 0.004)
6 of 8: p = 0.145
10
9 (p = 0.011)
10 (p = 0.001)
8 of 10: p = 0.055
12
10 (p = 0.019)
11 (p = 0.003)
9 of 12: p = 0.073
16
12 (p = 0.038)
14 (p = 0.002)
11 of 16: p = 0.105
20
15 (p = 0.021)
16 (p = 0.006)
14 of 20: p = 0.058
25
18 (p = 0.022)
19 (p = 0.007)
17 of 25: p = 0.054
How to run a fair comparison
Use the same source. Compare a lossless file with an MP3 or AAC encoded from that exact file, not two different downloads or masters (unless the masters are what you are testing).
Pick revealing material. Lossy artefacts show up most in cymbals, harpsichord, applause, castanets, sharp transients and quiet reverb tails. Loop a few seconds of such a passage.
Switch quickly. Auditory memory for fine detail fades within seconds, so short loops and fast A/X/B switching beat listening to whole songs.
Keep the volume comfortable and fixed. Use good headphones in a quiet room, and don't change the volume during the test.
Take breaks. Fatigue makes everything sound the same after 15 to 20 minutes.
Be honest with the numbers. Decide the number of trials in advance and keep every run, not just the best one.
What the results usually show
Well-made modern encodes at high bitrates (for example LAME MP3 at 256 to 320 kbps, or AAC at 256 kbps) are very hard for most people to distinguish from the lossless original in controlled ABX tests, while lower bitrates such as 128 kbps MP3 are often detected on critical material. Your own result depends on your hearing, your equipment and the music, which is exactly why a personal ABX test is worth doing. To check whether a file is truly lossless in the first place, use the Lossless Audio Checker.
Frequently asked questions
What is an ABX test?
ABX is a double-blind listening test. You hear two known samples, A and B, and an unknown sample X that is randomly either A or B. You decide which one X is, many times in a row. If you cannot really hear a difference, you will be right about half the time, just like flipping a coin; a score well above half shows that you can.
How is the p-value calculated?
The p-value is the probability of getting at least your number of correct answers by pure guessing, from the binomial distribution with a 50% chance per trial: p = sum of C(n, i) / 2^n for i from your score up to n. For example, 12 of 16 correct gives p = 0.038, and 14 of 16 gives p = 0.002.
How many trials should I do?
At least 10, and 16 is a good standard. With 16 trials you need 12 correct for p < 0.05. Fewer trials make it impossible to reach significance even with a good ear (with 4 trials, even 4 of 4 gives only p = 0.0625), and many more trials cause listening fatigue.
Why does loudness matching matter?
Even a difference of a few tenths of a decibel makes one version sound fuller or clearer, and listeners then pick the louder one rather than hearing the real difference. The test measures the RMS level of both files and turns the louder one down to match, which is standard practice for fair comparisons.
Why do the files need to be aligned?
MP3 and AAC encoders add a short delay of silence at the start (about 1,105 samples, or 25 ms, for many MP3 encoders). If A and B are not aligned, switching makes the music jump, which reveals which is which. The tool finds the offset by cross-correlation and compensates for it; you can fine-tune it in milliseconds.
Can I stop early when I'm doing well?
You can restart at any time, but the p-value is only valid for a number of trials fixed in advance. Stopping when you happen to be ahead, or repeating the test until you get a good score, inflates the chance of a false positive. Decide the number of trials first and report every run.
Are my files uploaded?
No. Both files are decoded and played by your browser's Web Audio API on your own device. Nothing is sent to a server.
Are all the tools on FreeToolCabin tested for quality and accuracy?
Yes! Every tool on our site undergoes automated Q/A testing in real browser environments to ensure high precision, correct output, and zero errors. If you happen to spot a bug or have a suggestion, please contact us at [email protected] and our team will review it right away.