Can you hear a difference between two audio files? Run an ABX test
If you think two audio files sound different, try a blind ABX comparison before deciding which one you prefer. You hear known examples A and B, then identify whether an unknown X is A or B. Repeat the choice across randomized trials. The result answers a narrow question: could you identify these two files in this setup? It does not rank their quality, test your hearing, or settle how they sound on every device.
Choose files that make a fair comparison
Start with two versions of the same passage. Ideally, they come from the same source and cover the same duration, with the same silence at the beginning. A different master, edit, intro, channel layout, or offset gives away the answer without testing the change you care about. WaveLocus starts both decoded files from sample zero for each preview, but it does not automatically align their musical content. Check the files before drawing a conclusion.
The local-file ABX tool accepts two files your browser can decode. It shows the decoded duration, sample rate, and channel count. If those differ unexpectedly, inspect the files first. The Web Audio decodeAudioData documentation notes that decoded audio is resampled to the AudioContext's rate. That means the displayed decoded format is session information, not necessarily an untouched statement about the original file container.
Remove easy clues before starting
Set the physical volume low, then choose a comfortable level. Leave the browser, output device, connection, and system volume unchanged for the whole run. If one file is obviously louder, you may identify loudness rather than the difference you intended to test. WaveLocus's optional RMS match attenuates the louder decoded file. It does not perform perceptual loudness matching, equalization, or editing. You can switch matching off when the level difference itself is the question.
Do a quick sighted preview of A and B. If they contain different starts or obvious silence, stop and prepare better source files. Avoid watching filenames, waveforms, or metadata during the blind trials if those clues could influence your answer. Do not raise the volume just to chase a faint difference. A short, comfortable passage is easier to repeat consistently than a long session that becomes tiring.
Run a local ABX comparison
- Load A and B. Select the two local files in the WaveLocus ABX tool. The files are decoded in browser memory and are not uploaded.
- Choose matching deliberately. Leave RMS matching on if an accidental average-level difference would be a nuisance cue. Turn it off if you specifically want to test that level difference.
- Choose a trial count before answering. The tool offers 5, 10, or 16 randomized trials. More trials provide more evidence, but only if you can listen attentively under stable conditions.
- Preview freely within each trial. Play A, B, and X as needed. Each preview restarts from the beginning. Only one selected source is audible even though the two sources start together.
- Answer without looking for hidden cues. Choose whether X matches A or B. The tool randomizes X again for the next trial and reveals the answer history only after the run.
Read the result without overreading it
The tool reports the number of correct answers and an exact one-sided binomial p-value. In plain language, that p-value asks how often random 50/50 guesses would produce at least this many correct answers in the selected number of trials. A small value is evidence against that guessing model for this run. It is not the probability that one file is better, nor the probability that your conclusion is true. Repeating many runs until one looks impressive also changes how you should interpret the result.
When the test is the wrong tool
ABX compares files, not headphones or speakers that you physically swap between previews. The browser cannot blind a hardware change for you. If the suspected difference appears only when a cable, output, or device changes, use a controlled path comparison in guided diagnosis. If one file is broken, silent, or clipped, you do not need a blind test to document that failure. A simple playback reference may be clearer.
A successful identification does not certify the cause. Two files may differ in level, start position, master, encoding, or content. A near-chance result cannot prove that two formats always sound the same. The conclusion is tied to the chosen files, listener, playback path, trial conditions, and number of answers. Professional listening studies control more variables and often use formal designs such as those described in ITU-R BS.1534. A personal browser run is useful evidence, not a substitute for that study design.
Next step
If the result surprised you, inspect the two files for offset, duration, level, and source differences, then run one planned repeat with the same passage and setup. Keep both results, including the trial count and matching mode. For a problem that follows an output device rather than a file, continue through guided diagnosis and change one path element at a time.
Sources and method
This guide describes WaveLocus's local-file ABX workflow and exact chance calculation for randomized A/B identification. MDN's decodeAudioData reference supports the browser-decoding boundary. ITU-R BS.1534 is a formal subjective-audio assessment method, cited for context rather than as a claim that this browser test follows its protocol. The test cannot grade audio quality, establish universal audibility, or turn a single personal run into a population study.