What does a curve tell you?
A frequency-response curve shows the measured level at different frequencies. The left is bass; the right is treble. A higher line means more measured energy at that frequency, not better sound.
The chart uses a logarithmic frequency axis, from 20 Hz to 20 kHz. Equal distances represent equal frequency ratios. Values between measured points are shown on a common grid; missing coverage stays missing.
Compare like with like.
Matching measures response distance: the weighted RMS difference in measured shape, in dB. Lower means closer on the rig, not a percentage match or a prediction of enjoyment. A weighted least-squares level offset removes differences in overall measurement level before comparison.
Balanced weighting gives less influence to uncertain high frequencies. Bass, vocal and treble focus change the weights, not the underlying measurements. Partition statistics inform reliability where supported; otherwise matching uses a conservative default profile, not measured error bars.
A product reference is the aligned, per-frequency median of its default measurements within one rig partition. Tunings are separate opt-in candidates, not quietly averaged into the default. Fit and unit variants remain inspectable evidence of variation. A per-reviewer distance range is the observed minimum and maximum distance to the reference, not a confidence interval.
711-family and B&K 5128 rigs are never pooled or ranked together; headphone rigs have separate partitions too. Compare can explicitly overlay different rigs, with a caution and rig labels. Its curves first receive the same balanced, reliability-weighted level offset to the first curve. The difference view then subtracts the selected baseline point by point, without an extra alignment.
Reviewer offsets, not rig conversion.
Similar and taste matching correct reviewer offsets by default only when partition support is sufficient and pooled mean improvement is positive overall and in every held-out fold. Otherwise correction stays off; explicit correction requests without sufficient support are rejected. Detail charts and EQ use uncorrected curves.
In the 2026-11 dataset build, five-fold validation on iem-711 reduced held-out same-product pair distance from 1.781 to 1.588 dB (−10.9%) across 50,823 pairs; fold improvements were 0.184–0.202 dB. This is not physical calibration, rig conversion or evidence of listener preference.
Precision is not certainty.
Seal, tips, insertion depth, unit variation and measurement setup can all change a curve. Upper-treble measurements deserve particular caution. Your ear and the measurement coupler are not the same acoustic system.
Product shading shows the minimum and maximum of the aligned default measurements when at least two sources contribute. It describes the available observations, not the range of all units or a statistical confidence interval. A single source does not establish certainty.
Frequency response alone does not describe distortion, spatial perception or every aspect of a listening experience. We do not turn it into an invented “sound quality” score.
The measurements belong in the story.
The catalog draws from public squig.link reviewer databases. Each measurement retains its source, rig, original files, variant and processing history. Product pages link to that evidence, and PNG exports retain curve attribution. Hardware and reviewer-listed USD prices are not independently verified.
Public availability is not a redistribution licence. Unless explicitly granted, permission is “public, not explicitly licensed.” Attribution is not a substitute for permission. We publish resampled curves, not raw text files. The site is non-commercial and credits every source. Any reviewer who prefers not to be included is removed through an exclusion and a rebuild. For correction, exclusion or takedown, open a GitHub issue.
Loading dataset sources…
Reference targets
Full definitions: methodology and maths · data sources and rights.
Your taste stays yours.
There is no account and no tracking. Curve analysis and EQ fitting run in your browser’s worker. Uploaded measurement text is never sent to a server. Theme preference stays locally; comparison product and measurement IDs, channel, smoothing and baseline can be shared in the URL.
Product or rig-matched reference-target EQ fits a constrained 48 kHz filter response. Its before/after distance is the level-aligned RMS residual over 20 Hz–10 kHz. Notches and high frequencies are treated conservatively. This approximates the target curve on the same rig, not its sound in your ear.
Back to Discover