ABX Blind Test Player: Run an ABX Test Online
Overview
ABX Blind Test Player. You feed the ABX Blind Test Player two audio sources — generated tones or your own uploaded files, entered as Source A and Source B — then it plays an unlabeled sample and asks you to click X was Source A or X was Source B across repeated trials. You'll get a real statistical significance score at the end, since guessing alone only gets you to 50%. Use the live spectrum meter to get an instant reading without any extra hardware.
Can you really hear the difference between two versions of a song, or is it expectation doing the listening? The ABX blind test player takes the guesswork out: you compare two reference clips, a hidden third clip plays, and your score shows whether you beat random chance. Every round takes under a minute, and the result arrives with its own statistics, so you know what the number means.
How an ABX Test Works
In an ABX test you get two references, A and B, plus a mystery clip called X. X is secretly a copy of A or B, picked at random each round, and your job is to say which one it matches. Because the answer stays hidden until you commit, price tags, brand loyalty and expectation cannot sway you. Only what reaches your ears counts, and that is why audiophiles and codec designers trust the format. The free dac and sound card quality test runs entirely in your browser via the Web Audio API — nothing is uploaded anywhere.
Sample X, Trials and the Number of Tests
Each round is one trial. A single trial proves almost nothing, because you have a 50% shot at the right answer by pure luck. Sample X changes every round, so you cannot memorise a pattern; you have to listen. Choose the number of tests before you begin (12 to 20 rounds is a sensible range) and never stop early because you happen to be ahead. Each time you vote, the player locks in your answer and marks it correct or wrong only when the session ends. Running multiple trials is what turns a hunch into evidence.
Keeping Your Listening Test Honest
A fair session depends on removing every clue other than the audible difference itself. Before you start, check these points:
- Match the loudness. A clip that is even 1 dB louder tends to win, which is bias, not a real difference.
- Check the sync. If the clips are offset, switching creates a click that gives the answer away.
- Skip the cheating. Reading file names or spectrograms defeats the purpose, and the result only fools you.
Using the ABX Blind Test Player
The player runs on the Web Audio engine in your page, so nothing needs installing and the playback you hear is decoded locally. Pick two versions of one recording, each a short sample, load a file or a streaming source, then follow the steps below. Everything you need is on screen: two reference buttons, a third for X, and two buttons to submit your answer. The frequency sweep test for speakers and headphones gives you a clear answer instead of guessing.
Loading Audio Files in Your Browser
Use your browser to open the two clips you want to compare, and keep them as similar as possible apart from the one thing you are testing. Trim each one to a 30-second sample of the same passage, and match the sample rate, since resampling alone can add an audible change.
- Load the high-quality audio version first, then the low-quality audio version.
- Set your trial count and put on headphones or switch to your usual speakers.
- Listen to A, B and X as often as you like, then vote for the one X matches.
Lossless vs Lossy Sound Quality
The most common reason to run this check is to see whether lossless files really beat lossy ones. A lossless FLAC file keeps every bit of the source, while a lossy encoder such as AAC or Opus discards detail that compression judges inaudible. At 320 kbps most listeners struggle to separate lossy from lossless, which makes music a hard case worth testing honestly. Drop to 128 kbps and the gap in sound quality often becomes much easier to catch. Your own equipment and ears decide where the difference becomes audible.
Play A, Play B and Hotkeys
You can switch between clips mid-playback, so the music continues from the same point and you compare the same moment each time. Learn the controls before the first trial:
- Play A / Play B
- Starts that reference; pressing the same button again stops it.
- X button
- Starts the hidden clip you have to identify.
- Hotkeys
- Keyboard shortcuts keep your hands off the mouse, which avoids distracting clicks.
Settling a 192 kbps vs FLAC Question With an ABX Tester
I want to know whether I can stop storing FLAC copies of my jazz albums, because they eat a 64 GB phone. The decision is simple: if I cannot tell a 192 kbps MP3 from the lossless original, I convert the lot. So I load one 30-second passage from a live recording in both versions, pick 20 trials, and put on my usual closed-back headphones.
After the session the player reports 14 of 20 correct, which is 70%. That feels like a clear win, but I check it against the cut-off the way the formula says: the combinations for 14 through 20 add up to \(38760 + 15504 + 4845 + 1140 + 190 + 20 + 1 = 60460\), and \(60460 \div 1048576 \approx 0.058\). A 5.8% chance by guessing sits just above the conventional 5% line, so the honest reading is "not proven".
| Scenario | Score | Chance by luck | Verdict |
|---|---|---|---|
| First session | 14 of 20 | 5.8% | Not proven |
| One more correct answer | 15 of 20 | 2.1% | Proven |
| Second session added | 27 of 40 | 1.9% | Proven |
One answer is the whole difference, which is exactly why I do not trust a single 20-trial run. I repeat the session the next morning, when my ears are fresh, and score 13 of 20, so the combined 27 of 40 lands at 1.9%. The number I would watch is the combined total, not either session alone.
The caveat is that this proves I can hear something on this passage; it says nothing about the rest of the library. Cymbal-heavy live tracks are where lossy encoders struggle most, so my verdict does not carry over to quiet piano pieces. My next step is concrete: keep FLAC for the live recordings, rerun the same 20 trials on one studio album, and convert it only if that run stays above 5%.
Reading Results From an ABX Tester
When the last round ends, the results panel reports how many answers were right and the percentage you scored. A high percentage alone is not enough; what matters is how unlikely that score would be if you had pressed buttons at random. That gap between your score and luck is the real difference you detected.
P-Value, Chance and Guessing
The p-value is the probability of scoring at least this well by guessing. Random picking here behaves like a coin flip, and the likelihood of a given score follows a binomial curve:
$$p = \frac{\sum_{k=c}^{n}\binom{n}{k}}{2^{n}}$$Here \(n\) is the number of rounds and \(c\) is how many you got right. Suppose you answer 12 of 16 correctly: the combinations add up to \(1820 + 560 + 120 + 16 + 1 = 2517\), and \(2517 \div 65536 \approx 0.038\). That is a 3.8% chance of reaching 12 or more by luck, below the usual 5% cut-off, so the statistical evidence says you heard something. The table shows how many correct answers you need at common session lengths.
| Trials | Correct needed (p < 0.05) | Chance of that score by luck |
|---|---|---|
| 10 | 9 | 1.1% |
| 12 | 10 | 1.9% |
| 16 | 12 | 3.8% |
| 20 | 15 | 2.1% |
Treat the table as a floor, not a target: reaching it with only a handful of rounds gives little confidence, because a lucky streak looks the same as skill.
Bonferroni Correction for Multiple Tracks
Testing several tracks multiplies the odds of a false positive, since each one carries its own 5% risk. The Bonferroni correction fixes that by dividing the cut-off by the number of tracks. For six tracks the new bar is \(0.05 \div 6 \approx 0.0083\), so a score only counts for an individual track at roughly 0.8% or lower. Passing that stricter statistical bar needs more rounds per track, which is why serious sessions run 20 trials each.
A claim you can only make with your eyes open is a claim about the label, not the sound.
Run the session, trust the numbers, and let the data tell you whether the extra bitrate or gear earns its place.
Frequently Asked Questions
- What is an ABX test?
- An ABX test gives you two references, A and B, plus a hidden clip X that is secretly a copy of one of them. You decide which one X matches, over several trials, so expectation and labels cannot influence the answer.
- How many trials do I need for a reliable result?
- More trials give a firmer answer. Around 10 to 20 trials per comparison is common, and you need at least 9 of 10 or 15 of 20 correct to beat the usual 5% cut-off.
- What does the chance by guessing number mean?
- It is the probability of scoring this well or better if you were picking at random, like flipping a coin each round. A figure below 5% is the usual sign that you really heard a difference.
- Why is the cut-off lower when I test several tracks?
- Each track carries its own 5% risk of a false positive, so the risk adds up across tracks. The Bonferroni correction divides the cut-off by the number of tracks, for example 0.05 divided by 5 tracks is 1%.
- What is the difference between an ABX test and an A/B preference test?
- An ABX test asks whether X matches A or B, which tests if you can hear any difference. An A/B preference test asks which sample sounds better, which tests whether your preference is consistent.
- Can I cheat or skip trials?
- You can, but the result then says nothing about your hearing. Fix the number of trials before you start, match the loudness of both clips and do not stop when you are ahead.
- Do lossless files really sound different from lossy ones?
- Often only at lower bitrates. Many listeners cannot separate high-bitrate lossy files from lossless ones, which is exactly what a blind test can show for your own ears and equipment.