Draft finder: where the window leaks air

Because a hand finds the room and an instrument finds the gap. Air moving through a small opening is turbulent, and turbulence radiates noise across a very wide band; up above 14 kHz an ordinary room has almost nothing else in it, so a leak that is inaudible stands far above its surroundings there. Measured on synthesised rooms: 33 dB above a quiet room in that band, against less than one decibel of change in the audible part of the same recording. The other half is wavelength — about twenty millimeters, short enough not to bend round the frame, so the reading peaks where the gap is rather than telling you that the window is draughty somewhere.

That is the question the check answers, and the reason it exists. Before scanning, the page plays a ladder of tones through the phone’s own speaker and listens for them coming back. If they come back at 19 kHz, the microphone hears 19 kHz and a silent wall really is a silent wall. If nothing comes back, the result is inconclusive rather than negative: the speaker, the volume and the microphone are all in that loop and this test cannot say which one gave out. In that case the two-phone mode is the way round it, because the source is the other phone.

Because an absolute number here would be fiction. Microphone sensitivity above 14 kHz varies enormously between phones, none of them are calibrated, and the background in a flat by a main road is nothing like the background in a house in the country. What can be measured honestly is a difference: this spot against a solid piece of wall two meters away, in the same minute, with the same phone. That is what the reading is, and it is also what you actually want to know.

Probably a tone rather than a hiss, and the page labels which. Turbulence through a gap is broadband — energy spread across the whole band. A fan blade, a phone charger, an LED driver, an old monitor and an ultrasonic pest repeller all ring at one frequency instead. The page measures how tonal the sound is and writes "tone" or "hiss" next to each mark, so the charger on the windowsill does not end up in the report as a leak.

One phone plays a tone at 18.5 kHz and switches it on and off seven times a second; the other looks not for energy at 18.5 kHz but for that seven-hertz pattern in what it hears. Anything else sitting on the same frequency — a television, a repeller, a switching supply — is steady, so it contributes nothing to the pattern. Measured on the fixture: an interferer 25 dB louder at the carrier still loses to the real emitter by 49 dB. Put the source inside, walk the frame outside, and the loudest reading is where the sound is getting through.

That is the easier case, and the one hardware detectors are sold for. A pressurised leak is far louder and far higher in frequency than a draughty window, so it stands out immediately — but the same limit applies: a phone microphone stops somewhere around 20 kHz while a purpose-built detector works at 40 kHz, where a workshop is quieter still. Close to the fitting, this will find it. Across a noisy workshop full of machinery, the 40 kHz instrument is worth its price.

No, and the gap is worth naming. A hardware detector listens at 40 kHz where there is even less to compete with, uses a directional horn, and is calibrated. A phone is limited to about 20 kHz by its own sampling, has an uncalibrated microphone with an unknown response up there, and points nowhere in particular. What it does have is the two things that matter for a window: a quiet band and a short wavelength. For finding which corner of a frame leaks, it works; for an industrial survey, it is not the instrument.