A transparency claim rests on a documented listening method

The tool moment

When a plugin, codec, or AI-assisted process is advertised as transparent, or indistinguishable from the original, that claim usually traces back to a specific listening-test design rather than a casual comparison. The International Telecommunication Union's Radio Communication Sector publishes Recommendation ITU-R BS.1116, titled Methods for the subjective assessment of small impairments in audio systems. Its version record shows the current revision, BS.1116-3, was approved in February 2015, following earlier versions dating back to 1994. The recommendation exists for the case where a difference between a reference and a processed signal is expected to be subtle, exactly the kind of claim a musician hears when a vendor says a stem-separation artifact or an AI mastering pass is not audible.

What the documents show

BS.1116 does not measure any specific product. It specifies a listening-test protocol: the recommendation's own text states that data on small impairments must come from listeners with demonstrated expertise in detecting them, because untrained ears are unreliable at this scale of difference. The prescribed design is a double-blind triple-stimulus method with a hidden reference, in which a known reference is labeled A while the hidden reference and the object under test are randomly assigned to B and C, and the listener grades any perceived impairment on a continuous five-grade scale. This differs from Recommendation ITU-R BS.1534, which the ITU's own documentation says is built for medium and large impairments rather than the small ones BS.1116 addresses. As a living recommendation subject to revision, this description reflects the document as retrieved on 16 September 2026.

What stays with the musician

The standard defines a method for detecting whether a difference is audible under demanding conditions; it does not decide whether that difference matters to a given mix, genre, or listener. A musician still has to judge whether a barely detectable artifact is disqualifying for a particular use, or whether an easily audible one is acceptable for the sound they are after. The standard does not certify any company's marketing claim; a vendor citing transparency has not thereby stated that a BS.1116-style test was performed at all.

Judge it by listening

Editorially, when a production tool claims to be transparent, it is worth asking what test, if any, backs that word, since the term carries a specific technical meaning under BS.1116 that an untrained, sighted comparison does not satisfy.

• Was a transparent claim tested with expert listeners and a hidden reference, or is it a marketing adjective?
• Is the impairment being judged small, BS.1116's territory, or more obvious, MUSHRA's territory?
• Would the same judgment hold up in a blind comparison, without knowing which file is which?

BS.1116 will not settle an argument about whether an AI-assisted tool sounds as good as the original recording. It offers a vocabulary for asking the question properly: who listened, whether they knew what they were hearing, and how the impairment was scored.