Online Blind ABX Test – FLAC or MP3?

Can you really hear the difference between a lossless file and a high-bitrate MP3? This free online ABX test lets you load your own two tracks, matches their level, and challenges you to identify a hidden sample — then tells you, with a statistical p-value, whether your result beats pure chance. It runs entirely in your browser and nothing is uploaded.

It exists because impressions lie. The moment you know which file is «the good one», you hear it as better, even when the two are identical. A blind, level-matched test removes both traps — you don't know which is which, and neither is louder — so what is left is your ears. Bring a FLAC and an MP3, two DACs, two EQ settings, or use the built-in training mode with a single file.

How to do the ABX test online, step by step

1. Load track A and track B. They must be the same musical excerpt in the two versions you compare. Cut the same 10–30 seconds from each; nothing is uploaded, the files are decoded in your browser (WAV, FLAC, MP3, M4A, OGG).
2. Keep level matching on. The tool equalizes the perceived loudness of A and B so a stray half-decibel doesn't decide the test for you. Leave it enabled unless you are deliberately testing a level difference.
3. Pick the number of trials (10, 16 or 20). More trials means a stronger result but more listening; 16 is the sensible default.
4. Run the trials. In each one you can switch freely between A, B and the hidden X — by button or with the keyboard (A / B / X, space to play or pause, 1 for «X = A», 2 for «X = B»). Switching is seamless and keeps its place, so you compare the same moment.
5. Decide and read the p-value. After the last trial you get your score, the exact p-value, a plain-language verdict, a running history of your sessions in this browser, and a shareable card.

No second file? Use the training mode

If you only have one track, the tool can generate track B from A with a known, controlled difference so you learn what a real positive feels like: an obvious +3 dB EQ bump at 3 kHz (easy), a subtle +1 dB one (hard), −4 dB of treble from 8 kHz, or a +0.5 dB level change. Start easy, confirm you can pass it, then work down to the subtle settings — it recalibrates your sense of what «I can hear it» actually means. To train your ears with controlled signals you can also play pure tones and noise with the online tone generator.

How to read the result: the p-value

The p-value is the probability that your score happened by luck. Below 0.05 (5 %) is the usual bar for saying the difference is audible to you. Guessing alone gets you half right, so the question is always «how far above half, and over how many trials?».

The tool computes the exact binomial p-value — P(X ≥ correct) with a 50 % chance per trial — so you never have to. Notice how few extra hits it takes: 12 of 16 already clears the bar, but 8 of 16 is exactly what guessing predicts. Running more sessions on the same material tightens the cumulative figure, which the history block keeps for you.

Why blind and level-matched

These two controls are the whole point; without them an ABX test is just a listening session with a scoreboard. «Blind» kills expectation bias — the pull of the price tag, the brand, the format name. «Level-matched» kills the oldest confound in audio: the slightly louder of two clips almost always sounds «clearer» or «fuller», and differences as small as 0.5 dB are enough to fool you.

This tool matches perceived loudness, not raw level, using K-weighting (the basis of the LUFS standard: it lifts the highs and filters subsonics to measure the way the ear hears). That is more honest than a flat RMS match, because two clips with the same RMS but different tonal balance are not equally loud to you.

What to use an ABX test for

FLAC vs MP3, and where bitrate stops mattering

The classic use: is a 320 kbps MP3 really distinguishable from the lossless original? For most music and most people the honest answer is «rarely, and only on revealing passages» — sharp transients, cymbals, reverb tails. This is where you find your threshold instead of repeating forum wisdom.

Streaming tiers and Bluetooth codecs

Now that Spotify, Apple Music and the rest ship lossless, the real question is whether you can hear it on your chain. Export the same clip at two tiers and ABX them. It matters most over Bluetooth: a lossy codec (SBC, AAC) versus a high-bitrate one (aptX HD, LDAC) is a far bigger, more audible gap than lossless-versus-320 usually is.

DAC, amplifier or cables?

This is where blind testing earns its reputation. Record or capture the same passage through two DACs or two amps, level-match, and ABX. Competent modern gear is often indistinguishable, which is a useful, wallet-saving result — and cables between line-level components essentially never survive a proper blind test. If the honest verdict is «I can't tell», that is real information, not a failure.

Before and after an EQ or a tweak

Made a small EQ change and not sure it did anything? Render the before and after and ABX them to see if the tweak is audible at all before you commit to it. It pairs well with checking how a headphone actually measures on the headphone frequency-response comparator.

What an ABX test proves — and what it doesn't

A significant result means you can tell A from B, on your gear, in these conditions. That is exactly what most arguments need and nothing more. It does not prove two files are bit-identical, and — crucially — a null result (you couldn't tell) does not prove they are the same. «No evidence of a difference» is not «evidence of no difference»: maybe the gap is real but below your threshold today, on this system, with this clip. Report it honestly as what it is.

How to set up a fair test

• Same excerpt, cut identically. A misaligned start or a different length turns a quality test into a timing test.
• Short clips work better. 10–30 seconds you can loop and re-hear beats a whole song; auditory memory is short.
• Use revealing material. Cymbals, sibilance, reverb tails and quiet decays expose codec artefacts; a dense wall of sound hides them.
• Good headphones, quiet room. You can't ABX a difference your chain or the background noise is already masking. Know your own ceiling with the online hearing test.
• Enough trials, and take breaks. Fatigue flattens your scores; 16 focused trials beat 40 tired ones.

If you suspect one of the two files is not what it claims to be, run it through the fake lossless detector first: it reads the spectrum and tells you whether it really came from a lossless source.

Frequently asked questions

A blind, level-matched ABX test shows whether a difference is audible to you on your gear — not whether two formats are bit-identical.