Decibel Meter

Press start and this reads the level from your microphone, names what that level sounds like, and tracks the peak, average and L10 across the session. You can export the whole log as a CSV, and there is an offset control so the scale can be made to agree with a real meter if you have one.

How to use it

1. Press Start and allow microphone access. Nothing is recorded — the signal is analysed frame by frame in the page and discarded.
2. Hold the phone at arm’s length, pointing towards whatever you are measuring. A hand cupped over the microphone can cost 10 dB.
3. Let it run. A single instant tells you very little; thirty seconds of a room gives you an average and an L10 worth reading.
4. Use Reset to clear the statistics without stopping, which is how you measure a second room.
5. Export to CSV when you want a record with timestamps.

Why no browser can give you a true decibel reading

This is the part every competing tool leaves out, and it is worth being direct about because it changes how you should use the number.

A decibel of sound pressure level is an absolute measurement, referenced to 20 micropascals — the quietest pressure variation a healthy young ear can detect. To report it you need to know how many pascals of pressure produced the electrical signal you are looking at. That requires knowing the microphone’s sensitivity in millivolts per pascal, and the gain applied everywhere between the microphone and your code.

A browser knows neither. The Web Audio API hands over a stream of floating point samples between −1 and +1, and that is the whole of the information available. The operating system has already applied an unknown amount of automatic gain, which differs between an iPhone 13 and a Pixel 8 and can differ between two units of the same model. There is no API that reports it.

What a browser can do is measure the signal accurately in relative terms. The root-mean-square of the sample window is a genuine measure of energy, and doubling the sound pressure genuinely produces a 6 dB rise in it. So the shape of the measurement is right; only its absolute position on the scale is unknown. Adding a fixed offset — 94 dB by default here — puts a typical phone’s readings roughly where they belong.

That is why the offset control exists rather than being hidden. If you have access to a real meter, set both going in the same room and adjust until they agree. From then on the readings are meaningful for your specific handset. Without that step, treat the numbers as good for comparison and rough for absolutes: this room is 15 dB quieter than that one is reliable; this room is exactly 47 dB is not.

Two further limits are worth knowing. Most phone microphones compress above roughly 100 dB, so a concert or a power tool reads lower than it is — the loudest environments are exactly where a phone is least trustworthy. And no consumer microphone applies A-weighting, the frequency curve that matches human hearing sensitivity and which every occupational measurement uses. Readings here are unweighted, so bass-heavy noise reads higher than a proper dB(A) figure would.

What it does not do

It is not a calibrated instrument and cannot be used for a legal noise complaint, an occupational exposure assessment, or anything where the number carries consequences. It does not apply A-weighting. It cannot measure below roughly 30 dB, because that is where the microphone’s own electrical noise floor sits. The session log is capped and the page must stay open — switching tabs stops the stream, deliberately, to release the microphone.

Nothing is uploaded. No audio is written to disk or transmitted.

What each level sounds like

Safe exposure time, NIOSH 3 dB rule