10 dB or 10 dB(A)? The difference that decides whether your noise problem gets solved

What a decibel is: a ratio, not a count

A decibel is not a ruler measurement. It is a ratio on a logarithmic scale, comparing the sound pressure you measured against a reference value of 20 µPa. The scale exists because the range the human ear copes with is enormous — from barely audible to painful is a factor of about a million in pressure.

• An increase of 3 dB is double the sound energy, yet most listeners barely register it as louder.
• An increase or decrease of 10 dB is roughly a doubling or halving of perceived loudness: about twice as loud, or about half as loud.
• A difference of 1 dB is effectively indistinguishable on a real site, and is usually of the same order as the measurement uncertainty itself.

Decibels therefore do not add directly. Two sources of 60 dB each do not make 120 dB — together they make about 63 dB. The reverse matters more in practice: if you silence the loudest source completely but a second source only 3 dB quieter remains, the total you measure falls by just a few decibels.

Seen from the other side, “a 10 dB reduction” is not a small number. In engineering terms it halves the perceived loudness. There is only one condition, and it is the one most often overlooked: the reduction has to happen in the sound that is actually bothering you, not in whichever number the supplier is free to pick.

Why A-weighting exists, and what dB(A) tells you

The human ear is not equally sensitive at every frequency. At ordinary everyday levels it is most sensitive in the mid range, roughly 1–4 kHz, and far less sensitive as frequency falls. A tone at 63 Hz needs to carry tens of decibels more energy than a tone at 1000 Hz before a listener judges the two equally loud.

A-weighting is a filter that imitates that behaviour. It discounts low-frequency energy according to a curve defined in IEC 61672, then sums all frequencies into a single number, written dB(A), meaning the whole sound approximately as a person hears it. This is why most environmental and occupational noise law worldwide uses dB(A) as its primary indicator.

• dB(A) is a summary figure: a single number describing the entire sound across all frequencies, after A-weighting has been applied.
• A dB value at a stated frequency, such as “10 dB at 1000 Hz”, describes that one band only and says nothing whatever about the other bands.
• A bare dB figure with neither a weighting nor a frequency stated has almost no meaning in a contract, because it can always be read in favour of whoever wrote it.

The summary figure and the band figures are therefore not interchangeable, and an improvement in one does not imply an improvement in the other.

The trap: “a guarantee of not less than 10 dB reduction”

Wording like this is common in turnkey noise-treatment quotations. Skimmed, it reads solid. Read as an engineer would, it fails to state the three things that matter — at what frequency, under what weighting, and measured where. Where those are not stated, the right to choose all three passes to the supplier by default.

The noise people actually complain about is usually low-frequency dominated

Most industrial and building noise complaints are not about shrill sound. They are about the hum and the thump that carry into a neighbour's home: fans and air handling units, compressors, chillers and cooling towers, transformers, standby generators, and distant traffic rumble. The bulk of their energy sits in roughly the 31.5–250 Hz range, where the wavelengths are long — so the sound diffracts around barriers, attenuates less with distance than higher frequencies, and couples into building structure readily enough that occupants sometimes feel it more than they hear it.

But most treatments perform best in the mid and high frequencies

Porous absorptive materials — glass wool, mineral wool, acoustic foam — have absorption coefficients that fall away sharply as frequency drops, because their performance depends on thickness relative to wavelength. Absorbing low frequencies effectively demands depth, or an air gap behind the material, on a scale the available space on site rarely allows. Partitions behave the same way: under the mass law, sound insulation rises with frequency and with mass per unit area, so a lightweight partition is intrinsically poor at low frequencies, and partition systems also have structural resonances and a coincidence dip that depress performance in specific bands.

The result: a true 10 dB that solves nothing

Put those two facts together and the mechanism is plain. A supplier who is entirely honest and delivers exactly what was promised can achieve a 10 dB reduction at 1000 Hz, evidence it with real measurements, and be in breach of nothing. But if the sound driving the complaint sits at 63–125 Hz, where that same treatment achieves only a few decibels, the overall A-weighted level at the receptor may barely move. What the complainant experiences is a change in the character of the sound — less sharp, duller — while the thump that keeps them awake at night is still there.

This needs no intent to deceive. It happens by default whenever the contract does not say which number it is talking about, and the whole of that risk sits with the buyer. Having the acoustic clauses and performance guarantees reviewed before you sign, by someone with no stake in the installation, is usually far cheaper than fixing the outcome afterwards.

How to read a noise guarantee

An acoustic performance clause that can actually be enforced answers five questions. If any one of them is missing, the guarantee is closer to marketing copy than to an obligation.

Model wording

“Not less than 10 dB(A) reduction in the overall A-weighted level at [named receptor], measured to [named standard and method], verified after completion by an independent party.”

That single line closes all five gaps at once, and two further points should always be added. First, state who pays for re-measurement if the result fails. Second, set an absolute target alongside the reduction figure, because a 10 dB(A) reduction from a very high starting point can still leave the receptor above the applicable criterion. If you are not sure what level you have to reach, check the table of noise limits applied in Thailand before agreeing to a number in the contract.

Where dB(A) is, and is not, the right measure

Describing dB(A) as “the sound as people hear it” is fair as far as it goes, but stopping there hides an important limitation. The A-weighting curve was derived from equal-loudness contours at fairly low levels, yet it is applied at all levels. The consequence is that it discounts low-frequency energy too heavily for sound that is loud and bass-rich — so some of the noise residents find most intolerable is exactly the noise a dB(A) figure under-represents.

• Low-frequency hum and thump from fans, compressors, chillers or transformers, disturbing sleep at night
• Bass from entertainment venues or sound systems, transmitted through building structure into neighbouring dwellings
• Sound with a pronounced tonal component, which annoys more than its dB(A) level suggests
• Very low frequency noise and infrasound, which falls outside the range A-weighting treats meaningfully

In these cases better instruments already exist, and they belong in the specification from the outset.

• The difference between dB(C) and dB(A) at the same position is a sound first-pass screen: a large difference indicates dominant low-frequency energy, and a warning not to judge the case on a dB(A) figure alone
• Octave or one-third octave band criteria assessed against NR or NC curves, which is the most direct approach for requirements inside buildings
• Assessment and penalty adjustment for tonal or impulsive character, following the guidance in the ISO 1996-2 series

The right question before signing is therefore not “how many dB do you guarantee”, but “at what frequency does our problem sit, and does this guarantee address that frequency?”. Answering it takes a frequency spectrum measurement at the receptor before the remedial works are designed.

FAQ

In Thailand, is dB(A) always the legal measure?

It is the primary one, but not the only one. Thailand's general noise standards are expressed in A-weighted terms, both as a level averaged over a stated period and as a maximum level, and the occupational noise requirements use dB(A) as well. What actually binds you, however, may come from elsewhere: mitigation measures set out in a project's environmental impact assessment, the rules of a condominium juristic person, or the terms of a private contract, any of which may impose band-limited or stricter criteria. Nuisance findings under the public health law are also not tied to a single number in every case.

Why does the number in my complaint differ from the supplier's number?

Usually because nobody is lying — the two parties measured different things and gave them the same name. The common differences are position, time of day, averaging period, weighting, and index: one side reports an energy average over a period while the other reports the maximum of an event.

The other frequent cause is comparing a laboratory test value for a material directly with a field measurement. Laboratory figures come from a perfectly installed specimen under controlled conditions, whereas a real building always has flanking transmission, junctions, service penetrations and leaks. Field performance is therefore lower than the catalogue value as a matter of course. The correct approach is to place the two documents side by side and ask, line by line, which line differs.

Can I check a guarantee myself?

You can do a useful first pass, and you should do it every time. Work through whether the wording names all five items from the section above: the quantity, the weighting or frequency, the measurement position, the standard and method, and who verifies after completion. If any is missing, ask for it to be written in before you sign — amending wording before signature costs almost nothing, while amending it after installation is usually impossible. What is hard to judge on your own is whether the guaranteed figure is enough for your problem, because that requires spectrum data for the source and the level at the receptor, not document review alone.