MUSHRA¶

A multiple-stimulus audio-quality test using a visible reference, hidden reference, impaired anchors, and listener ratings for intermediate impairments.

Version: v1 · 2026-09-28 · History
Domain-specific #: 10864
Domain group: Applied Sciences & Engineering
Origin domain: Engineering & Design (beyond software)
Subdomains: Audio Engineering, Listening Tests → Engineering & Design (beyond software)
Aliases: Multiple Stimuli with Hidden Reference and Anchor

Core Idea¶

MUSHRA—Multiple Stimuli with Hidden Reference and Anchor—is a controlled way to ask listeners about the perceived quality of several processed versions of an audio excerpt. The listener sees the original reference, rates candidate stimuli, and also encounters an unlabeled copy of the original and deliberately impaired anchors. These extra items reveal scale use and help interpret ratings; they are not candidate codecs whose engineering performance is inferred directly.

The frozen account locates MUSHRA in intermediate audio-quality assessment and cites an ITU-R recommendation for its versioned details. A 0–100 listener rating can distinguish audible degradations under a common comparison frame, but it is conditional on program material and listener population. More sensitive tests may be preferable for subtle impairments, and expert listeners may weigh artifacts differently from ordinary users. The entry states the evaluative structure, not a timeless screening threshold or a ready-made test protocol.

Structural Signature¶

Sig role-phrases:

Audio reference — Provides the original signal whose perceived quality anchors comparison. It is constitutive. Counterfactual: Without an original reference the task changes into another opinion-rating design.
Candidate stimuli — Supply several coded or processed alternatives rated by the same listener in one comparative frame. It is constitutive. Counterfactual: One isolated item does not instantiate the multiple-stimulus design.
Hidden reference — Repeats the original without its label as an internal quality and reliability check. It is constitutive. Counterfactual: Removing it discards a named element of MUSHRA's design.
Impaired anchors — Place recognizable lower-quality items on the rating scale to stabilize its interpretation. It is constitutive. Counterfactual: Without anchor signals the named hidden-reference-and-anchor methodology is incomplete.
Listener judgment and frame — Records quality ratings under the test's listener, scale, and intended intermediate-impairment scope. It is boundary. Counterfactual: One score cannot be read as an objective codec property independent of material and listeners.

What It Is Not¶

A single MOS rating. One sample judged without the comparative references and anchors is another listening design.
Objective bit-rate measurement. Codec size and computational cost are not identical to perceived audio quality.
A guarantee of tiny-defect sensitivity. Other tests may suit nearly imperceptible differences better.
One universal quality score. Audio material, listener training, and presentation conditions bound the inference.
Closest near-miss. A high-fidelity A/B discrimination study may compare audio samples rigorously but belongs to another protocol if it lacks MUSHRA's hidden reference and anchors.

Scope of Application¶

Audio-codec comparison. Compare several processed versions against one original excerpt.
Listening-study interpretation. Use hidden reference and anchors to understand listener scale behavior.
Method selection. Distinguish intermediate-quality assessment from more sensitive small-impairment testing.
Result reporting. State listener cohort and program material before generalizing ratings.

Clarity¶

Find the visible original, multiple processed samples, hidden original, impaired anchors, and listeners' graded quality judgments. A low score for the hidden reference signals a reliability problem, but exact exclusion thresholds depend on the cited protocol version. Do not call an unreferenced single-sample score MUSHRA or infer codec quality beyond the material and listeners studied.

Manages Complexity¶

The method puts several audio alternatives into one listener's common reference frame rather than treating independent scores as directly interchangeable. Hidden and impaired controls make calibration visible, but the resulting number still compresses listener variation, program material, and artifact type.

Abstract Reasoning¶

Identify the original excerpt and the candidate processed signals.
Check that the same listener can compare multiple stimuli against the visible reference.
Locate the hidden copy of the original and the degraded anchors.
Read scores in relation to the listener cohort and intermediate-impairment aim.
Keep version-specific screening and tiny-defect claims separate from the method's structural identity.

Knowledge Transfer¶

The reference–candidates–hidden-control–anchor–rating arrangement transfers among codec and audio-use studies when comparable excerpts and listeners are specified. A particular ITU edition's numeric screening rule, a music result, or an expert panel's sensitivity does not transfer unchanged to all content or listeners.

Examples¶

Canonical¶

Listeners compare several lossy-coded renditions of one music excerpt alongside the visible original, an unlabeled copy of it, and audibly degraded anchors. Scores express perceived intermediate quality relative to that common excerpt.

Mapped back: Audio reference → visible original music excerpt; Candidate stimuli → several codec renditions; Hidden reference → unlabeled duplicate original; Impaired anchors → recognizably degraded versions; Listener judgment and frame → 0–100 quality ratings for that excerpt and cohort.

Applied / In Practice¶

A panel rates one audio codec sample on a five-point opinion scale with no original, hidden copy, or impaired anchors. The result may be useful listening evidence but is not the MUSHRA design.

Mapped back: Audio reference → absent; Candidate stimuli → single item; Hidden reference → absent; Impaired anchors → absent; Listener judgment and frame → ordinary opinion rating.

Structural Tensions¶

T1 — Common Reference versus Listener-Specific Judgment. A shared original and anchors make ratings comparable, but listener expertise and audio content still affect what is heard.

Diagnostic: Which quality differences and listener population do these scores represent?

T2 — Intermediate Sensitivity versus Tiny Impairment Detection. MUSHRA resolves substantial quality differences efficiently, while very small defects may require another more sensitive test.

Diagnostic: Are the impairments in the method's intended range?

Structural–Framed Character¶

The approved DAG parent is Evaluation: bounded audio outputs receive listener quality ratings under explicit criteria. MUSHRA adds a visible original, hidden original, impaired anchors, and a graded scale for intermediate audio quality.

Evaluative weight: Explicit, but scores depend on listeners, excerpts, and test design. Human-practice-bound: High, because auditory judgment and calibration controls constitute the method. Institutional origin: ITU methodology standardizes variants; edition-specific screening rules should not be universalized. Vocabulary travels: Codec studies can reuse the design after matching content and listeners. Import versus recognize: Recognize MUSHRA by its candidate/reference/control arrangement; an ordinary preference poll imports no calibrated comparison.

Its character: A controlled audio-evaluation subtype with portable comparative scoring and specific stimulus controls.

Structural Core vs. Domain Accent¶

Skeletal core. Bounded objects are judged against a criterion frame through observations and scores.

Domain-bound accent. Processed audio, visible original, hidden original, degraded anchors, listeners, and intermediate-quality ratings define MUSHRA.

Why not prime. Evaluation is broader; without these audio controls the test is a different assessment.

This entry is a kind of Evaluation.

Strict parent — evaluation. The bounded audio outputs are compared with a criterion-bearing reference and anchors to yield quality scores; generic evaluation need not use hidden copies or listening.
Related — mean opinion score. Both collect judgments, but MOS does not by itself entail simultaneous reference/hidden-reference/anchor comparison.

Relationships to Other Abstractions¶

Current abstraction MUSHRA Domain-specific

Parents (1) — more general patterns this builds on

MUSHRA is a kind of Evaluation Prime

MUSHRA scores bounded audio outputs against a visible original and calibrated hidden/anchor controls under a listener-quality criterion.

Hierarchy path (1) — routes to 1 parentless root

MUSHRA → Evaluation → Comparison → Self Checking

Neighborhood in Abstraction Space¶

MUSHRA sits in a moderately populated region (40^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Communication, Learning & Information Practices (15 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

MOS. Tell: Are hidden reference and degraded anchors present?
Codec throughput. Tell: Is the output a perceptual rating rather than file size or speed?
ITU revision. Tell: Which published protocol edition supplies a numerical rule?
Very small impairment. Tell: Is the intended difference below MUSHRA's stated sweet spot?

References¶

Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/MUSHRA (revision 1363949308).
Preserved source candidate: http://www.itu.int/rec/R-REC-BS.1534/en
Preserved source candidate: http://www.itu.int/rec/R-REC-BS.1116/en
Preserved source candidate: https://www.itu.int/rec/R-REC-BS.1116-3-201502-I/en
Preserved source candidate: http://asa.scitation.org/doi/abs/10.1121/1.1904305
Preserved source candidate: http://www.aes.org/e-lib/browse.cfm?elib=15471
Preserved source candidate: http://www.aes.org/e-lib/browse.cfm?elib=17177
Preserved source candidate: http://www.aes.org/e-lib/online/browse.cfm?elib=16088
Preserved source candidate: http://www.aes.org/e-lib/browse.cfm?elib=16089

The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.