Online vs Hardware Decibel Meter

The honest comparison is not close on any axis that involves the number itself. A dedicated sound level meter wins on every measure of what the figure is worth. A browser wins on the two things that decide whether anyone takes a reading at all: it is already installed, and it costs nothing.

Which of those matters depends entirely on what you are going to do with the result, so this page is organised around that rather than around a feature table.

What hardware actually gives you

A microphone whose behaviour is known. An instrument is built around a measurement microphone with a published frequency response, chosen for flatness rather than for making speech intelligible. Everything downstream depends on that, and it is the one thing no software can supply.

Something to check it against. An instrument is designed to be calibrated, and to hold that calibration between checks. Research meters are used exactly this way — when Kardous and Shaw needed ground truth for their app comparison, they verified levels with a type 1 sound level meterKardous & Shaw. The reference existed because there is a class of device built to be one.

Headroom. Phone inputs run out somewhere around 100 to 105 dB. That is below the level of the events people most want a figure for, and past it the number stops climbing while the room keeps going.

A result someone else has to accept. This is the real difference. NIOSH's position is that where a sound level meter is not available, sound measurement apps can provide a measure of area noise but may not comply with regulatory requirementsNIOSH. An employer, an inspector, a court and an insurer are all downstream of that sentence.

What a browser gives you that hardware does not

It is already there. The reading you take is worth more than the reading you would have taken if you owned the right equipment. Most of the noise questions people actually have — is this room worse than that one, is the flat louder since the new tenants, does the extractor fan help — are never asked of an instrument because there is no instrument in the room.

It costs nothing to try. Not a small point when the alternative starts in the low hundreds and the question might turn out not to need an answer.

It leaves no trace. Audio on this site is read in the browser, turned into a number and discarded. Nothing is recorded and nothing is uploaded.

It is fast enough to be repeated. A figure you can take nightly for a fortnight is a different kind of evidence from one careful reading, and the repetition is often where the answer actually lives.

Three questions that decide it

Does anyone else have to accept the number? If yes — a landlord, an employer, a council, an insurer, a court — you need hardware, and the argument ends there. A browser reading is not going to survive contact with someone motivated to dispute it, and neither is a calibrated phone.

Do you need the absolute level, or the difference? Most of the error in a browser reading is systematic: the same device is wrong in a consistent direction, so the error largely cancels between two readings taken the same way on the same device. If your question is comparative, a browser answers it about as well as an instrument would. If you need the absolute figure, the browser meter accuracy page sets out how wide the band around it is, and it is wide.

How loud is the thing? Below the point where a phone input runs out of headroom, a browser is reading a signal it can actually see. At a loud venue it may not be, and the failure is silent — the number simply stops responding. That is the range where hardware stops being a nicety.

How often will you need it? A one-off question — is this restaurant loud, is the new washing machine worse than the old one — is a browser question, and buying an instrument for it is a way of not answering it this week. A recurring one is different. If you are checking a workshop every month, documenting a neighbour over a season, or making decisions about a room you are responsible for, the case for hardware compounds: the readings become comparable with each other over a long period, the device does not change underneath you, and the figures are worth something to a third party at the end of it.

The awkward middle case is the recurring question you did not know was going to recur. That is worth starting in a browser and moving, because the browser readings are still useful as a before — as long as you noted how you took them.

Where the two overlap more than expected

Both need the same settings to mean the same thing. NIOSH names the settings it recommends for a sound level meter: a 3 dB exchange rate, A frequency weighting, and Slow responseNIOSH. Those are the settings this site's meter runs, so the number it produces is the same kind of number, described the same way, even though it is produced by much worse hardware.

A hardware meter set to C weighting and Fast response is not directly comparable with a browser reading in dBA either, which is a mistake people make in both directions. Which weighting to use and what each one changes is dBA vs dBC.

Both also benefit from being checked against a reference, and the phone benefits more. Setting a calibration offset does not close the gap, but it does remove the constant part of the error, which on most devices is the largest part.

What Class 1 and Class 2 mean, roughly

Instruments come in classes, with tighter tolerances at the top. The study above used a type 1 meter as its reference for exactly that reasonKardous & Shaw.

This site is not going to print the tolerance limits for each class, because doing that properly means quoting a standard it does not have in front of it, and half-quoting a standard is worse than not mentioning it. The practical shape is enough for a buying decision: the higher class is what you want if the figure will be challenged, the lower class is what most people buy, and both are in a different category from anything running on a phone. Neither this page nor the meter it belongs to conforms to IEC 61672 in any class.

What to do if you only have a browser

Use it for what it is good at and be explicit about the rest.

Take the reading the same way every time: same spot, same height, same device, same orientation, and let it settle rather than reading the first number. Write down what you did alongside what you got. Prefer a series to a single figure — five nights at the same hour says something a single Tuesday cannot.

And say what it is when you report it. "An uncalibrated browser estimate, roughly 78 dBA, taken from the same spot at 11pm on five consecutive nights" is a sentence that survives scrutiny, because it does not claim more than it can support — and NIOSH's own line about apps not necessarily complying with regulatory requirements is the reason the qualifier belongsNIOSH. "78 decibels" on its own invites exactly the challenge it cannot answer.

The free online decibel meter on this site is built for that use: it draws the uncertainty as a band rather than a needle, warns when the input is clipping or the browser has refused to switch off automatic gain, and exports what it recorded so the series is the evidence rather than your memory of it.