Sound Quality Judging Criteria: What SQ Competition Actually Scores
Sound quality judging is often described as subjective, and in the loosest sense it is — the instrument is a trained listener in a driver's seat. But a defensible SQ result is not an opinion poll. It is a structured evaluation against written criteria, performed at a fixed seat, with a fixed track list, at a controlled reference level, and recorded on a score sheet that another judge can audit. This article sets out what those criteria actually measure, why measured frequency response and perceived quality diverge, and how a rubric converts listening into a repeatable score.
Ask ten enthusiasts what makes a car sound good and you get ten answers. Ask ten judges working from one rubric and the answers converge — not because they agree on taste, but because the rubric fixes what to listen for, on what material, and how to turn it into a number. That translation is the technical problem of SQ judging.
The four measurement domains behind an SQ score
Most credible rubrics decompose into four families of criteria. They are not independent, but they fail in distinguishable ways — and separating them lets a judge explain a score instead of asserting it.
1. Tonal accuracy
Tonal accuracy asks whether instruments and voices have the timbre the recording intends. A judge audits frequency balance across bands: whether the low end has weight without masking the midrange, whether male and female vocals sit in the same tonal family, whether cymbals have air without sibilance. Because hearing sensitivity is not flat, tonal judgements are level-dependent — equal-loudness behaviour means the same system sounds bass-shy and treble-shy played quietly. Hence a fixed reference playback level rather than volume set by feel.
2. Staging and imaging
Staging concerns the apparent size, height, width and depth of the reproduced scene. Imaging concerns whether individual sources within it are focused and correctly placed — a vocal that is a point, centred and stable rather than a smear drifting toward the louder door. In a car the obstacle is geometry: the driver sits far off-axis, so the path to the near speaker is much shorter than to the far one. Sound travels roughly 34 cm per millisecond, so a 34 cm path difference is a 1 ms arrival difference — enough for the precedence effect to pull the image hard toward the near side. Per-channel delay and level trim correct it; judging means hearing whether the correction holds across the frequency range or only where the installer tuned.
3. Linearity under level
A system that sounds composed at moderate level and falls apart when pushed is not a good system, and a rubric that evaluates one volume setting will never catch it. Linearity criteria ask what changes as level rises: does tonal balance shift, does the centre image collapse, does bass become boomy as cabin modes are driven harder, do voice coils heat and compress dynamics, does clipping add audible harmonic and intermodulation products. Distortion audibility does not track total harmonic distortion in any simple way — low-order harmonics can be nearly inaudible at percentages that look alarming on a datasheet, while a little high-order or intermodulation distortion is immediately obvious as hardness — one reason listening tests survive alongside instrumented measurement.
4. Noise floor and system hygiene
The noise floor is what the system produces when the programme material produces nothing. With the vehicle running and volume at the reference position, a judge listens for broadband hiss, alternator whine appearing as tones that track engine speed, ground-loop hum, and switching artefacts. Noise is inseparable from gain staging: amplifier gains set far higher than the source requires amplify the system's own input noise along with the music. Because dB is a logarithmic ratio, small numbers matter — a noise floor 10 dB higher is ten times the noise power.
How a written rubric removes judge bias
Bias in SQ judging is rarely malice. It is drift: a judge's internal reference moves over a long day, and the third car gets scored against the second rather than the standard. A rubric attacks drift structurally rather than relying on judge discipline.
- 1Anchored scales. Each band has a written description of what earns it, so a 6 means the same thing in the first car and the twentieth. Scales without anchors compress toward the middle.
- 2Fixed evaluation material. A defined track list with a stated reason for each track — centre image stability, low-frequency control, vocal timbre, mix separation.
- 3Fixed seat and fixed level. A stated seating position, a reference level set from a known signal, and a rule about whether the engine runs. Change any of these and scores stop being comparable.
- 4Independent scoring before discussion — comparing notes first collapses two opinions into one.
- 5Deviation review. Where two judges differ by more than a defined margin, the sheet is reviewed rather than silently averaged; persistent divergence usually means a badly worded criterion.
- 6Written comments tied to criteria. Notes attached to the line they justify, which makes a score feedback rather than a verdict.
Illustrative rubric structure
Weightings differ by format and class; treat the table as structure, not a published scale.
| Criterion | What the judge listens for | Typical failure mode |
|---|---|---|
| Tonal accuracy | Band-to-band balance; timbre of voices and instruments at reference level | Midrange dip that flatters vocals on one track, hollows them on the next |
| Staging | Stage width, height and depth; whether it sits ahead of the windscreen | Stage collapsing into the doors; height pinned at knee level |
| Imaging | Focus and stability of sources; centre lock across frequency | Centre vocal pulled toward the near door above 2 kHz |
| Linearity | Tonal and spatial stability as level rises; onset of compression or clipping | Balance and image intact at moderate level, both lost when pushed |
| Noise floor | Residual hiss, engine-correlated tones, hum, switching artefacts | Broadband hiss from gains set well above what the source requires |
Measured response versus perceived quality
A microphone at the headrest and a listener in the seat do not report the same thing. Understanding why is the difference between tuning a car and tuning a graph.
- The cabin is a pressure chamber at low frequency. Below its lowest resonant mode, response rises as frequency falls — which is why a car produces usable output well below what the same driver manages in free air, and why a flat measured curve down to 30 Hz often sounds bass-heavy.
- Steady-state magnitude response hides time. An averaged curve says nothing about arrival times, so two systems with near-identical response can image completely differently.
- Early reflections colour timbre. Reflections arriving within a couple of milliseconds of the direct sound sum and cancel with it, producing comb filtering a microphone captures literally and hearing partly discounts.
- Perceived loudness is not measured level. A 10 dB increase is tenfold the acoustic power but only roughly a doubling of perceived loudness, so level-matching by ear and by meter are different operations.
Target curves: useful tools, not verdicts
A target curve is a response shape a tuner aims for at the listening position, typically with a gently rising low-frequency shelf and a controlled high-frequency roll-off rather than a ruler-flat line. Curves of this family exist because listener-preference research consistently finds a smooth, slightly bass-tilted balance preferred over literal flat in-room response. Two cautions follow. A target curve shortens tuning but is not a scoring criterion — the curve is only a proxy for the perception being scored. And it means nothing without its measurement conditions: smoothing, averaging and microphone positions change the trace enough that the same system can appear to hit or miss the same target.
The measurement tells you what the system did. The rubric tells you what the listener got. Report only one and the result stops being useful to the competitor.
What this means for competitors
The implications are dull but firm. Fix gain structure before touching equalisation, because you cannot equalise a noise floor away. Set time alignment and levels before shaping response, since correction applied to a misaligned system bakes the misalignment in. Verify behaviour at more than one volume, because linearity is scored. And treat the score sheet as the deliverable: criterion-linked written feedback is how competition improves builds.
Frequently asked questions
Is sound quality judging subjective?
The instrument is a trained listener, but the process is not an opinion poll. A written rubric fixes the seat, the reference level, the track list and the scale anchors, so two judges evaluating the same car are answering the same defined questions rather than reporting general preference.
Why do judges use a fixed reference level?
Human hearing sensitivity varies with level, so the same system sounds bass-shy and treble-shy when played quietly and fuller when played louder. Fixing the playback level keeps tonal judgements comparable between cars and across a long judging day.
Can a car measure flat and still score poorly?
Yes. A steady-state frequency response curve contains no timing information, so a car can measure smoothly and still image badly because of arrival-time differences, early reflections or a collapsing stage. It can also measure flat and sound bass-light, since cabin gain means a literally flat low end is rarely the preferred balance.
Should a system be tuned to a target curve?
A target curve is a useful starting point that shortens tuning, not a scoring criterion. Judges score the perceived result, and any curve is only meaningful alongside its measurement conditions — microphone positions, smoothing and averaging all change whether a system appears to hit it.
