Blind Listening Tests
A blind test needs three things, no labels, matched levels, random order.
Related
The problem
Blind tests usually fail because the test is invalid, not because the listener cannot hear. You know which file is AI, which is yours, which is version two, and the judgement is already bent. Add a one or two dB level difference and the louder file wins almost every time. One pair, one listen, one conclusion produces preference, not difference.
Why blind
Listening depends on context. A visible label sends you looking for evidence of the answer you already hold, which is confirmation bias. Level differences also masquerade as quality, since louder seems clearer and more detailed. A blind test only has to do three things, remove labels, match level, randomize order. Then the answer is about sound rather than about who you want to win.
Practice
Cut fifteen to thirty seconds from the same section with the same start point and length, and match integrated loudness rather than peak. Rename files A and B, let someone else shuffle the order, and write answers down before discussing them. Run at least ten trials per pair, not one; fewer than eight wins out of ten means no audible difference. Fix the question in advance, whether it is which structure is clearer or which you prefer, and never mix the two. Preference tests allow no preference; discrimination tests force a choice. Include a control with an obvious difference, such as a three dB high cut, to prove the setup works. Do not let the same person set the test and take it, or randomisation is wasted. Stop after twenty minutes, before the ears tire.
What to listen for
The live Chopin Ballade and the 1968 remaster are two performances, not two treatments of one take, so confirm that before testing. Bach's Toccata in D major and its remastered file differ by more than three minutes, which shows the label cannot be trusted and must be removed. Compare the same passage at the same level, not the words on the cover.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Sources
- 盲听与 ABX 方法的基本流程，依听测方法学通识核对，未抄原文
- 等响与电平偏差对判断的影响，依心理声学公开资料核对，未抄文本
Used for fact-checking only; all prose is original.