Antinodal
The honest version

What a phone can
measure, and what it can't.

Every room-acoustics app has to decide what to do about the gap between what it can measure and what you would like to know. Most of them fill it in. This is where we draw the line, and the measurements that put it there.

The rule

An uncalibrated microphone is bad at absolutes and remarkably good at comparisons. It has no idea whether it is hearing 74 dB or 81 dB. But ask it whether the left speaker reaches you before the right one, or whether this seat rings longer than that one, or whether the curtain you just hung changed anything — and it is comparing itself against itself, where the calibration it never had cancels out entirely.

So we report comparisons, and we say so. Every figure in the app carries what it rests on: the room's geometry, its materials, the microphone, or nothing at all.

Where the line falls

We can hear this

  • How long the room takes to go quiet, in the mid and high bands
  • Whether one speaker reaches your ears before the other, and by how far
  • Whether one side arrives brighter than the other — which is usually aim, not electronics
  • When the first reflections arrive, and from where
  • Whether what you changed last week helped

We can't hear this

  • Absolute sound pressure. A phone has no reference, so any dB figure would be invented
  • Bass below roughly 100 Hz — the microphone stops responding and the phone's own processing fights the measurement
  • What a surface is made of — the phone can't tell plaster from a curtain. What it can do is see exactly where each reflection lands, and time it

The second column is the interesting one, because it is where an app is tempted to draw. A decay chart with a solid block of colour under 300 Hz is not a measurement of your bass — it is a picture where the data ran out. We would rather leave the space empty and say why.

Evidence

Three rooms, one phone

The check that matters is not whether a number looks plausible. It is whether the numbers separate rooms that are genuinely different, in the order physics demands. A small room with a curtain and a wooden floor must be the driest; a tiled kitchen the liveliest; a furnished living room in between.

RoomTime to go quiet
Small room, curtain, wood floor0.27 s
Furnished living room0.35 s
Tiled kitchen0.47 s

Nearly two to one across the three, in the right order, from a phone speaker and a phone microphone. The living room was then measured again in a separate session and analysed independently: every frequency band landed within 0.02 to 0.04 seconds of the first result.

The band below 125 Hz correctly refused to answer in the larger room. That is not a gap in the table — it is the measurement declining to guess.

Evidence

Checked against a tape measure

Timing is the thing this method does best, so it is the thing we tested hardest — against the crudest and most trustworthy instrument there is.

Sitting midway between the two loudspeakers, seven measurements averaged +0.03 ms, which is the correct answer: nothing. Then the listener deliberately moved about 17 cm to the left, where a tape says the difference should be half a millisecond or so. The phone read +0.50 ms. Moving the same distance the other way reversed it.

One reading is good to about 15 cm of path difference, and that limit is the hand holding the phone rather than the arithmetic. We say so in the app, on the screen where the number appears, rather than in a disclaimer nobody reads.

Speaker-to-wall distances are a different case: both are measured in the same instant by the same sensor, so the shared error cancels and they resolve to under two centimetres — confirmed with a tape. Absolute distances round to five centimetres, because five centimetres is what repeats.

What we won't do

No score. Not a global one, not one per category. A room is not a number to be raised, and turning it into one turns listening into a game.

No promised decibels. An uncalibrated microphone cannot honour "this will give you 8 dB", so we say what you should notice instead.

No wall of findings. Five recommendations, ranked by how much they will change what you hear against how much work they are. Deciding what matters most is the job you came to us for; handing you thirty-eight problems is handing it back.

Nothing you can't check. Every reflection point comes with the two tape measurements that locate it, and the reason it is there: it is the one spot on that wall where a mirror would show you a speaker from where you sit. Go and hold a mirror there. We would rather be caught being wrong than believed for no reason.

For the curious

The test file

Antinodal can play its own test signal, but on a good system it is better to play ours through your own chain. This is that file: a sweep on the left channel, a pause, the same sweep on the right, repeated three times behind a silent lead-in so you have time to sit down.

It is an ordinary 48 kHz 16-bit stereo WAV. Put it wherever your music lives — a Roon library, a NAS, a USB stick — and play it like any other track.

↓ antinodal-sweep-v1.wav · 51 s · 9 MB

Play it at a normal listening level, not loud. It runs to 20 kHz and it is a full-scale signal.