It is not the volume. It is the words.

Most advice about noise treats it as a quantity: fewer decibels, better concentration. That is roughly true and it misses the thing that matters most, which is what the noise consists of.

Why speech is the expensive kind

The effect has a name — the irrelevant speech effect — and it is one of the more robustly replicated findings in this area. Background speech interferes with tasks involving verbal memory and comprehension considerably more than non-speech noise at the same level. Steady, broadband sound like ventilation or distant traffic is comparatively cheap. A single conversation you can make out the words of is expensive.

The reason is that language is not optional. You cannot decide not to parse speech in a language you know, in the way you can decide to ignore a fan. It arrives already processed, and it competes for the same machinery you are using to hold a sentence in your head.

This has a practical consequence that runs against intuition: a moderately noisy room can be easier to work in than a quiet one with one audible conversation. Intelligibility, not level, is what to reduce. A room where speech is present but not decipherable is measurably better than a room where one voice carries.

The level at which it starts to matter

Somewhere around 55 dB is the level commonly cited as the point where sustained attention starts to degrade noticeably — roughly the level of a lively conversation a few metres away. It is not a cliff. It is the region where the cost stops being negligible, and it is worth knowing because ordinary open-plan offices sit right on it and classrooms frequently exceed it.

In our own recorded sessions, evening work above 62 dB lost focus roughly twice as fast as quieter sessions. That is our data on our users and we would not present it as a general law — but the direction matches the literature closely enough to act on.

What actually helps, roughly in order

• Distance from the source, more than volume. Sound falls off with distance and intelligibility falls faster. Moving a desk away from a doorway, a corridor or a coffee point usually beats any attempt to make the whole room quieter, and it costs one afternoon.
• Masking, not silence. Steady broadband sound raises the floor so speech stops being decipherable. This is why a fan, rain, or a well-designed masking system often works better than noise-cancelling headphones in a room with one talker — you are lowering intelligibility rather than level.
• Music without words, if any. Lyrics reintroduce exactly the problem you are trying to solve. The evidence for instrumental music improving concentration is genuinely weak, but the evidence that lyrics interfere with verbal work is not.
• Soft surfaces. Carpet, curtains, upholstered panels. They shorten reverberation, which is what makes distant speech carry across a hard-floored room.
• Headphones, with a caveat. Active cancellation is very good at steady low-frequency sound — engines, ventilation — and much less good at speech, which is the part that costs you. They help on a train more than they help in an office.

Children, and why classrooms are different

Children are not small adults for this. Speech perception in noise develops through childhood, so the same background level costs a younger listener more than an adult. Anyone learning in a second language pays more again, and so does any child with a hearing difference, glue ear, or a language disorder — and there is usually at least one in a class of thirty.

The result is that a room the teacher experiences as workable can be genuinely hard for a handful of pupils, and it is not the same handful every day. That asymmetry is the case for measuring the room rather than polling it. What we build for classrooms shows a teacher the conditions — noise, air, light — and how the class as a whole is holding attention, with nothing about any individual child.

Measuring it without recording anyone

A microphone in a room is a sensitive object, and it should be. Ours measures level only: it does not record, transmit or store audio, and it is not built to be able to. What comes out is a number in decibels sampled through the day, which is enough to say the room passed 55 dB for two hours this afternoon and nothing at all about what was said in it.

Pair that with a measure of sustained attention and the question stops being an argument. At a desk, you find out whether it is the room or the hour — and if it turns out to be the hour, that is a different problem with a different fix. Stale air is usually worth ruling out at the same time, since it tends to arrive with the noise in any occupied room.