How the source data on this site was built

Who did this work. Reading 488 products' worth of manufacturer documents is not something one person does by hand in evenings, and this page would be dishonest if it did not say so: the finding, reading and transcribing of those documents was done with AI assistance. What is not AI is the rules below — what counts as a usable figure, what gets thrown away, what gets flagged rather than converted. Those are decisions, and they are the author's. The reason everything on this site is labelled, sourced and downloadable is that a method good at volume needs to be checkable by people who were not involved in it. If you find something wrong, the about page lists what has already been found and corrected, including in the author's own reasoning, and corrections are genuinely welcome.

Every amplifier, dongle and player on this site carries a power figure at 32, 150, 300 and 600 ohms. Almost no manufacturer publishes all four. This page explains exactly what was published, what was worked out, and what was thrown away — because a number you cannot trace is worth nothing, and a site that hides the difference is selling you something.

Of the 280 sources in the database, 270 have been traced back to a manufacturer document, and 10 are still carrying figures collected when the database was first assembled. Those 10 are being re-sourced one at a time, and until then their pages say so in plain words rather than quietly looking the same as the rest.

Published. The manufacturer states this power at this exact load. Nothing was done to it. 528 of the 1312 figures in the database are this.

Derived. The manufacturer publishes a figure at some other load, and this one was worked out from it. The working is always the same, and it is the reason this site exists: an amplifier is a voltage source with a current limit, not a power source. So the conversion runs through voltage. A published figure of 180 mW into 32 ohms is √(0.180 × 32) = 2.40 volts; that voltage is then held between published loads, and held constant beyond the highest one. Below the maker's lowest published load there is one further step: the voltage is clamped so that the implied output current never exceeds the highest current any published point implies, because current, not voltage, is what a real output runs out of first down there. That is why a figure below a maker's lowest published load comes out lower than holding voltage alone would give — on the Ferrum OOR, whose lowest published load is 60 ohms, holding voltage to 32 ohms would claim 3750 mW where this site prints 1067. Holding voltage constant beyond the top published load understates what the amplifier can really do, because its voltage usually rises a little into a higher impedance. The error therefore runs in the safe direction: this site will tell you an amplifier is short of power before it tells you one is enough.

Upper bound. 24 sources publish a maximum output voltage and no power rating at any load at all — Qudelix, Astell&Kern and Tanchjim all do this. Voltage into a load can only ever be lower than voltage with nothing attached, so a power figure computed from it is a ceiling, not a rating. Those pages say upper bound on every line, and the real figure into a low impedance may be a good deal lower. A few products go further and label the figure “Condition: No Load” in so many words, with nothing else on the sheet. Those are excluded rather than carried as a ceiling, because a voltage the maker has explicitly measured open-circuit says nothing about current at all, and the list below names them.

Through the output impedance. A few amplifiers — valve designs above all — publish one power figure and a high output impedance. For those the output impedance is not a footnote, it is the whole behaviour: it forms a voltage divider with the headphone, so the power an amplifier delivers peaks when the headphone impedance equals its own and falls away on both sides. The Bottlehead Crack, at a published 120 ohms, makes more power into 150 ohms than into either 32 or 600. That shape is computed from the two published numbers and nothing else.

A figure with no stated load is not a specification. Several well-known products publish exactly that — a bare wattage with nothing to attach it to — and they were left out rather than guessed at. So were products whose own documents contradict each other, and products whose figures could only be found on a dealer page that no two dealers agreed on.

Some of these will come back if the manufacturer publishes a load alongside the number, or if a factory document turns up that settles a conflict. None of them will come back as an estimate.

Sensitivity is where this gets worse, and it is the reason this site exists at all. A maker can publish it per milliwatt or per volt, and the two differ by 10 × log₁₀(1000/Z) — about 5 dB at 300 ohms and 18 dB at a 16 ohm in-ear. Read one as the other and a headphone looks far easier to drive than it is.

Every one of the 578 models here has now been checked against the manufacturer's own document. The result: 253 publish per milliwatt and needed nothing done; 139 publish against a voltage reference — per volt, or in a few cases an SPL at one stated drive voltage — and the figure on their page is that number converted with the working shown; 141 print a bare number with no reference at all; 33 publish no sensitivity figure whatsoever; 11 are electrostatics, which are specified on a different scale entirely; and 1, the Superlux HD 668B, carries a reference its own datasheet contradicts. Those six groups account for every model in the database.

Where a maker prints a bare number, the page shows both readings: what the figures mean if that number is per milliwatt, and what they would mean if it is per volt. The per-milliwatt reading is the one used in the tables, because it is the more common convention on a spec sheet that omits the unit — but the alternative is stated on the page, with the exact decibel gap, rather than hidden.

7Hz, AKG, Austrian Audio, Bowers & Wilkins, Campfire Audio, EPZ, Elysian Acoustic Labs, FiiO and 15 others publish per volt on at least one model. 7Hz, AFUL, Audio-Technica, Austrian Audio, Beyerdynamic, CCA, DITA Audio, Dunu and 29 others print at least one sensitivity with no reference at all — several makers appear in both lists, because the convention changes from model to model inside one catalogue: the spec sheet says “94 dB” and stops. There is no honest way to convert a number whose unit nobody states, so those pages say so and present the figure as an indication rather than a specification. Beyerdynamic quotes one 96 dB spanning every impedance version of the DT 770, 880 and 990, from 32 to 600 ohms. That is what a per-milliwatt figure looks like — efficiency barely moves when a coil is rewound — and not what a per-volt figure looks like. It is a good reason to think the number is per milliwatt, and still not a statement Beyerdynamic has made, so this site does not convert on it.

One maker was caught printing a reference its own product cannot accept. Superlux rates the HD 668B at 98 dB per watt — on a headphone whose own datasheet caps input at 300 milliwatts. Converted honestly that lands 30 dB below anything physical, so the page reports the contradiction rather than picking whichever reading looks better. Campfire Audio goes the other way and publishes an SPL at one specific drive voltage, 11.84 millivolts — unusual, but fully convertible, and converted.

Eleven headphones were researched for this database and then left out of it, because no sensitivity figure exists for them anywhere — not from the maker, and not in the material this database was first assembled from. That is the difference between them and the 33 models above, which also have no published figure but do carry an inherited one, and say so on their pages. They include the Grado Hemp (impedance published, the “SPL 1mW” row that every other Grado carries simply absent), the Focal Bathys in wired mode, the Sendy Audio Aiva, the Koss ESP/95X, the Sennheiser HD 25 Light (a maximum SPL is published, which is a different thing), the Tin HiFi T2, and the Etymotic ER4P and EVO. A model with no sensitivity cannot be put through any of the calculators here, and inventing one would defeat the point.

Three products were removed outright because they do not exist: a Limited Edition DT 1990 PRO, which Beyerdynamic has never made; a Focal Utopia in-ear, which Focal has never made; and a KZ ZA10, which is not a KZ model. All three were in the database, and all three were wrong.

111 sources have no output impedance traceable to the maker, and 197 have no traceable noise figure. On those pages the damping check and the hiss estimate are not run at all. It would have been easy to substitute a typical value and let every page look complete; the page says the number does not exist instead. The 10 oldest entries here are the awkward case: they carry both figures, but from independent measurement rather than from a document, so they are counted with the ones that published nothing and are marked unverified where they appear.

8 of the sources here have no manufacturer document behind any of their addresses either — the numbers come from a retailer, a review site or a manual aggregator. Those pages say so in the same words.

On the headphone side the same rule turned up 23 models whose figures stand on a shop's product page rather than on the maker's own site — Linsoul, HiFiGo, Concept Kart and a handful of others, across 10 brands, 7Hz, Thieaudio and Simgot worst affected. Shops usually copy the factory table faithfully, and most of these are probably exact. But a shop can carry an old revision, and nothing on this site would notice if it did, so each of those pages says so in its provenance section in as many words. It is the weakest provenance here, and the offer on every one of those pages is the same: find the figure on the maker's own site and the note comes off.

For most of them that offer is empty, and the pages now say so. 22 of the 23 name the reason on the model's own page, because the reason turned out to be the interesting part: 7Hz has no website at all — its own official link hub lists Instagram, Facebook and Threads, and twelve product links, every one of them to a shop. Kefine's domains do not resolve. BLON's do not resolve. HarmonicDyne's does, and serves a certificate issued for a different name, so no browser that checks certificates will open it. Thieaudio has a site with a “Technical Details” tab that renders empty and three of these four models returning 404. FiiO, NiceHCK and Tangzu each still run a site and have simply deleted the model from it.

That is worth saying plainly, because it is not what the spec-sheet-audit numbers suggest on their own. For a whole layer of this market the retailer is not a second-best source that a diligent person could improve on. It is the only place the figures have ever been printed, and every review, every database and every comparison you have read about these models traces back to the same shop page.

The other honest limitation is that a manufacturer's own figure is a claim, not a measurement. Where a maker quotes power at 1% distortion, that is a louder number than the same amplifier at 0.1%, and this site records the distortion condition where one was printed. Independent measurements exist and are better than any of this — they are just not something this site can reproduce.

If you find a figure here that disagrees with a manufacturer document, the document wins. The page footer has the contact address.