What does the p-value mean?

It’s the probability of scoring this well (or better) by pure guessing. For example, 12 of 16 correct gives p ≈ 0.038: guessing would do this well less than 4% of the time, so it’s usually taken as evidence you can hear a difference. A high p-value doesn’t prove the files sound identical, only that this test found no evidence; run more trials or pick a more revealing passage.

Why match loudness and align the files?

We hear slightly louder sound as clearer and better, and differences of 0.2 to 0.5 dB already bias results. MP3 and AAC encoders also add tens of milliseconds of delay, which would make the audio jump when you switch and give the answer away. The tool turns the louder file down to the same integrated loudness and aligns B to A, so only the quality difference remains.

How can I make the test more reliable?

Use good headphones in a quiet room. In sighted practice, find the few seconds where the difference is most obvious (cymbals, applause and reverb tails are common) and loop them. Turn off per-trial feedback for the real test, do at least 16 trials and repeat if needed. Don’t rely on the waveform or tiny switching noises.

Which formats work, and is anything uploaded?

Formats your browser can decode (WAV, FLAC, MP3, AAC/M4A, Ogg, Opus and more) are read directly; others are converted with a local copy of FFmpeg. Decoding, analysis, encoding and playback all happen in your browser, and files are never uploaded.

👂

👂ABX Blind Test - Can You Hear FLAC vs MP3? Online Audio Test

Can you really hear the difference between lossless and a 320 kbps MP3? The reliable way to find out is an ABX test: A and B are known, X is randomly one of them each time, and you decide which. This tool decodes both files at the same sample rate, measures integrated loudness per ITU-R BS.1770 and turns the louder file down to match, because even 0.5 dB louder tends to sound “better”. It then finds the time offset, such as encoder delay, by cross-correlation to the exact sample and aligns the two. During playback A and B run in sync and switching only changes the gain with a 10 ms crossfade, so the position never jumps and there are no clicks; drag on the waveform to loop the passage where differences are hardest to hear. Choose 8 to 40 trials. X stays hidden during the test, and at the end you see every answer and its timing, plus a one-sided binomial p-value and a verdict (with 16 trials you need at least 12 right for p < 0.05). There’s also a blind AB preference mode that randomly labels the files 1 and 2 on each trial and checks for a consistent preference with a two-sided test. No second file? Generate an MP3, AAC or Ogg Vorbis version of A at 32 to 320 kbps with FFmpeg. Everything runs in your browser; your audio is never uploaded.

Free · No sign-up · Runs in your browserLast updated: 2026-10-26

Loading tool...

How to use ABX Blind Listening Test

1Choose two audio files, or just A and generate an MP3 / AAC / Ogg Vorbis at a chosen bitrate as B
2Check the loudness match and alignment, practise sighted and drag a loop region on the waveform
3Pick ABX or blind AB preference and the number of trials, then read the p-value and verdict

FAQ

What does the p-value mean?: It’s the probability of scoring this well (or better) by pure guessing. For example, 12 of 16 correct gives p ≈ 0.038: guessing would do this well less than 4% of the time, so it’s usually taken as evidence you can hear a difference. A high p-value doesn’t prove the files sound identical, only that this test found no evidence; run more trials or pick a more revealing passage.
Why match loudness and align the files?: We hear slightly louder sound as clearer and better, and differences of 0.2 to 0.5 dB already bias results. MP3 and AAC encoders also add tens of milliseconds of delay, which would make the audio jump when you switch and give the answer away. The tool turns the louder file down to the same integrated loudness and aligns B to A, so only the quality difference remains.
How can I make the test more reliable?: Use good headphones in a quiet room. In sighted practice, find the few seconds where the difference is most obvious (cymbals, applause and reverb tails are common) and loop them. Turn off per-trial feedback for the real test, do at least 16 trials and repeat if needed. Don’t rely on the waveform or tiny switching noises.
Which formats work, and is anything uploaded?: Formats your browser can decode (WAV, FLAC, MP3, AAC/M4A, Ogg, Opus and more) are read directly; others are converted with a local copy of FFmpeg. Decoding, analysis, encoding and playback all happen in your browser, and files are never uploaded.