MUSHRA names the anchors a listening test cannot skip

The system

MUSHRA, short for “MUlti Stimulus test with Hidden Reference and Anchor,” is a standardised subjective listening-test method defined by the International Telecommunication Union's Radiocommunication Sector in Recommendation ITU-R BS.1534-3. As retrieved on 16 September 2026, the current text is that October 2015 revision, with earlier versions from 2001, 2003 and 2014 shown on the recommendation's own series page. It was built for comparing systems with medium-to-large audio impairments, such as codecs, and has since become a common method for comparing the perceived quality of generative-audio outputs as well.

What the documents establish

The recommendation itself specifies the test's structure in detail rather than leaving it to convention. A listener is presented, in each trial, with an open reference plus all systems under test, plus a hidden copy of that same reference and one or more hidden anchors, typically a low-quality and a mid-quality anchor; the document recommends “no more than 12 signals” per trial. It requires the test to be double-blind, lets the listener switch near-instantaneously between all stimuli, and converts each rating to “normalized scores in the range 0 to 100, where 0 corresponds to the bottom of the scale.” The hidden reference lets the analyst check whether a listener can reliably identify the undamaged original, and the anchors, the document explains, exist because “anchors can affect the test results” and “the name MUSHRA is often misused for tests not using reference and anchors” — the standard's own authors flag the mislabelling problem directly.

Craft and rights

For anyone reading a vendor's claim that its generated audio “won a MUSHRA test,” this document is the checklist: was there a genuinely hidden reference, a low-quality anchor, a panel blind to which system produced which stimulus, and a trial near the recommended signal count. None of that is a rights question in the legal sense, but it is a craft-and-evidence question in the same family: a listening-test result is only as trustworthy as its adherence to a named, checkable method, and BS.1534 is the document that defines what “MUSHRA” means.

Outcomes and open questions

The recommendation does not itself certify any particular vendor's implementation, and it explicitly separates its own scope from small-impairment testing (covered by the related Recommendation ITU-R BS.1116). It has been revised four times since 2001, so a cited MUSHRA methodology should be checked against the specific BS.1534 revision it claims to follow, since anchor and scoring guidance has shifted across versions. Readers should watch for informal “preference tests” being rebranded as MUSHRA results without the hidden-reference and anchor structure this document requires.

• Did the test include a genuinely hidden reference and at least one low-quality anchor, as BS.1534 specifies?
• Which revision of BS.1534 does a cited methodology claim to follow, and does its trial design match that revision's signal-count guidance?
• Is a vendor's internal listening test being described as “MUSHRA” without disclosing whether it followed the recommendation's blind, multi-stimulus structure?

BS.1534 turns “we ran a listening test” into a specific, falsifiable claim, and that specificity, not the brand name MUSHRA itself, is what should carry weight when a generative-audio result cites it.