ABX audio test
Before you start: Listen at a comfortable low volume. Files with different levels can make a louder version seem better. Stop between comparisons if listening becomes uncomfortable.
Blind comparison
Can you identify the difference?
Choose two audio files. WaveLocus decodes them only in this browser, starts each preview at sample zero, and makes only A, B, or randomized X audible. The files are not otherwise time-aligned, and they are never uploaded.
RMS matching measures average digital sample energy and only attenuates the louder file. It is not LUFS, perceptual loudness matching, or a claim that the files are otherwise equivalent.
Compare two local recordings by identifying whether a hidden randomized X matches A or B.
Interpretation: Correct identifications across the chosen trials can support a repeatable distinction under these listening conditions. The result includes an exact one-sided binomial chance calculation.
Limit: It does not establish which file sounds better, why they differ or a general hearing ability. Different passages, timing and levels can provide unintended clues.
Safety and privacy: Use a comfortable low listening level and take breaks. Files are decoded in this browser. Optional RMS matching attenuates the louder playback without editing either original file.
How to use this test
- Choose matching passages: Select File A and File B, then choose Load local files. Both previews start at sample zero, so align the passage in your editor first.
- Choose the comparison settings: Select RMS match or No match and a trial count. Keep the output path unchanged, then choose Start blind ABX.
- Identify X in each trial: Use Play A, Play B and Play X to compare, then answer A or B. Complete the selected trials and review the correct count and chance p-value without treating it as an audio-quality score.
Why ABX does not use a fixed test-tone level
ABX plays your files. RMS matching only attenuates the higher-energy file; it does not ensure equal perceived loudness. Use the same passage, channel layout and matching mode throughout the trial. Decoded overload or non-finite samples prevent a usable comparison. A result concerns these files and listening conditions, not a universal audio-quality score.
Source: ITU controlled-listening guidance. Read dBFS vs dB SPL to distinguish digital signal levels from physical sound pressure. Use the audio normalizer to prepare a deliberately level-adjusted copy before starting a separate comparison.
How to interpret your abx audio test results
The chance p-value is small: The calculation concerns the selected trials under a random-guessing model. It does not measure how large a difference is, prove its cause or establish which file sounds better.
The result is inconclusive: This does not prove that the files are identical. Trial count, the selected passage, matching and listening conditions affect the test.
Levels or passages differ: Align comparable material before a new session. RMS matching attenuates the higher-energy file; it is not guaranteed perceptual loudness matching.
Source: ITU controlled-listening guidance.
Frequently asked questions
What is an ABX audio test?
Compare known samples A and B with a hidden randomized X, then decide which file X matches. Repeated trials test identification under these conditions, not which recording is objectively better.
Are my comparison files uploaded?
The files are decoded locally. Optional RMS matching changes listening level, not the saved originals. Files start at sample zero, so choose matching passages yourself.
Does a low chance probability prove one file sounds better?
No. The result describes identification across the chosen randomized trials under these conditions. Level differences, mismatched passages or other cues can influence identification. It does not establish preference, audible quality or a general hearing ability.