How we test, and what we cannot test

We do not own the equipment we write about. That single fact shapes everything below, and stating it plainly is more useful than any claim we could make about a listening room.

What we do instead is measurable and checkable: we collect published specifications for a large set of products, convert them to comparable units, and report where a given model sits among its peers. A number that cannot be made comparable does not appear in a ranking on this site.

The sample

Figures above cover the products we have actually published a review of, and they move as the site grows. Specifications, prices and ratings are taken from retail listings and recorded with the date they were read; a price quoted in an article is the price on that date, not a live figure.

Why we normalize before comparing

Manufacturers publish the same quantity in different notations. Microphone sensitivity arrives as -32 dBV/Pa, as 2.7 mV/Pa, as 25 mV / 94 dB SPL, and sometimes as a bare -54 dB with no reference pressure named at all. These are not four different measurements. They are one measurement in four costumes.

Before any comparison we convert to a single unit. Three conversions do most of the work:

• Microphones: millivolts per pascal, from decibels relative to one volt.
• Headphones: sensitivity per milliwatt and per volt differ by an amount that depends on impedance — up to 15 dB between a 32-ohm and a 300-ohm model. Ignoring this makes one headphone look thirty times louder than it is.
• Speakers: 2.83 V equals one watt only into 8 ohms. A 4-ohm speaker quoted in volts appears 3 dB more sensitive than it is.

What we refuse to rank

A measurement without its conditions is not comparable, and we would rather publish an empty cell than a misleading one.

Frequency range without a tolerance. A range quoted where output has fallen 10 dB will always look wider than the same speaker quoted at 3 dB. One studio monitor in our set publishes both: 74 Hz at −3 dB and 54 Hz at −10 dB. Same speaker, 20 Hz of difference, chosen by which line of the datasheet gets printed.

Self-noise measured on different weighting curves. The ITU-R 468 scale reads 11–13 dB higher than A-weighting. One microphone in our set publishes both — 13 dB A-weighted, 24 dB CCIR — and comparing across the two would label a quiet microphone noisy.

Maximum SPL without a distortion threshold, and peak power of any kind. Peak wattage has no agreed definition and is not compared here.

Anything from a product listing that turned out to describe a bundle. Where a listing covers a microphone packaged with a shock mount, or a pair of speakers sold with an accessory kit, the specification fields belong to more than one object. Those listings are excluded rather than untangled.

Two failure modes we check for by rule

One name, two devices. A headset publishes Sensitivity twice — once for its microphone, once for its earpieces. Both numbers are plausible. Picking the wrong one produces headphones with a 10 kHz ceiling. We separate them by sign: headphone sensitivity is positive in dB SPL, microphone sensitivity is negative in dB relative to a volt.

One word, two quantities. Impedance in a headphone specification (17 to 600 ohms) and impedance in a speaker specification (4 to 8 ohms) are not the same measurement and are never placed in the same table. The same applies to the four distinct quantities that all appear as "power" in speaker listings: amplifier output, power handling, recommended amplifier power, and mains input.

Owner ratings

Where we report what owners think, we report the distribution — how many gave five stars, how many gave one — rather than the average alone. The average conceals whether a 4.6 came from broad satisfaction or from a split between delight and disappointment.

These ratings are aggregated from retail listings. We do not reproduce the text of anyone's review.

What specifications cannot tell you

Honesty about the limits of the method matters more than the method. Specifications describe compatibility and boundaries. They do not describe sound.

Not published by anyone, and therefore absent from this site unless we say otherwise: the shape of a frequency response curve, which is the reason one model sounds bright and another dull; distortion as a function of level; clamping force and fit, which is the usual reason headphones stop being worn; unit-to-unit variation; and what fails first, and after how long.

So our claims take the form "this model needs this much gain and withstands this much level", not "this model sounds better". The first is checkable. The second, from a specification sheet, would be invention.

Images

Product photographs come from the retailer listings named on the about page; the first image in each gallery names their owner. Charts and diagrams are ours, built from the dataset described above.

Corrections

If a number here is wrong, it is wrong in a way that can be demonstrated — that is the point of publishing the method. Write to us and it gets fixed, with the correction noted on the page.