Overnight Noise Monitor

You woke at three. By morning you cannot prove a thing — not when it started, not how long it went on, not whether it was worse than Tuesday. That gap between what you heard and what you can show is where most noise complaints die.

This page keeps the record for you. It reads the microphone once a second all night and writes down a single number each time: the level for that second. Nothing else. In the morning there is a chart of the whole night, a list of every stretch that crossed the line you set, and two ways to take it away.

What it writes down

One level per second, from the moment you press start. Eight hours is 28,800 numbers, which is far too many to read, so the page does two things with them.

It draws them, so you can see the shape of the night — where the quiet baseline sat, when it lifted, whether the bad half hour was one event or twenty. The chart's axis is a clock rather than a stopwatch, because every question anyone will ask you is in o'clock.

And it lists the loud stretches. Any run of seconds at or above your warning mark becomes an entry with the time it began, how long it ran, and the highest second inside it. Two runs a second apart are one entry, not two — otherwise a noisy hour becomes four hundred rows nobody will read. The mark is yours to set. It starts at the published line below which sound is unlikely to damage hearing however long it lasts, 70 dBANIDCD, and it lives in the settings under the chart next to the curve and the correction. Those are the same settings the meter on this site uses, saved on this device rather than sent anywhere.

Read the night by when, not by how loud

The single figure people reach for is the loudest moment, and it is the weakest thing in the record. It is one second, it might have been you shutting a drawer, and it invites an argument about whether a phone can be trusted to report it.

The strong part is the pattern. Something that starts at 23:40 every weeknight and runs until two is a pattern. Something that happens once is an incident. Whoever you take this to — a landlord, a managing agent, a council officer — is deciding whether to act on a pattern, and the list of times is what shows one. Click any entry and the chart centres on it, so you can see what the minutes either side looked like.

Zoom out to the whole night to see the shape; zoom into a minute to see a single event. Drag to move through time. The Live control keeps the plot following the newest second while the night is still running, and turns itself off as soon as you go looking through what has already happened.

What the figures mean, and what they do not

The number this page writes each second is a sound-level estimate from a device microphone: an uncalibrated one, and not what an environmental health officer will put in a report. Phone and browser sound apps have been found to scatter by roughly ±10 dB against a reference instrumentKardous & Shaw, and that error follows your device all night — which is exactly why the record is worth having anyway. The gap between a quiet hour and a loud one is real even when neither number is, because the same unknown error sits under both.

So read it as a relative record of when it got louder, not as a figure anyone will quote back at you.

For the thresholds themselves, three published numbers are worth holding in mind, and only one of them is about hearing.

A 1974 EPA report, summarised by NIOSH, named 55 dBA outdoors and 45 dBA indoors as the levels above which noise starts to interfere with speech and become an annoyanceNIOSH. Those were never issued as a standard and nobody enforces themNIOSH, but they are the right order of magnitude for a bedroom, and a night that sits in the fifties indoors is a night worth writing down.

Sound at or below 70 dBA is unlikely to damage hearing however long it goes onNIDCD. Nearly every domestic noise problem lives entirely below that line. Say so yourself before someone says it to you: the case is about sleep, not about damage.

85 dBA is where NIOSH sets its recommended limit for an eight-hour working dayNIOSH. It is drawn on the chart because it is the one line on the site that a reading can be compared against without argument, and it is an occupational figure — it says nothing about whether a flat is livable. A room can be miserable at 55 dBA and never come near it.

There are published night-time guidelines for environmental noise, and this page does not print one. They are written as long-run indicators taken outside the building, they are addressed to the people planning roads and airports, and setting one beside a single night's reading taken indoors on an unknown microphone would be borrowing authority the reading has not got. The rule that actually binds is local, and it is usually written as a time of night rather than a decibel figure.

What a phone will do

It will stop, and sooner than you think. Every mobile browser suspends the microphone when the screen locks, and most of them throttle a background tab long before that. A phone left on the nightstand does not record a night; it records until the display times out.

Use a laptop on mains power instead, with this tab open and in front, and the lid up. If the machine will sleep on its own, press Keep the screen awake — where the browser supports it, the page asks the system to hold the display on, and asks again every time you come back to the tab. Where the browser does not support it, the button is not shown, because a button that quietly does nothing is worse than none.

The page notices when it was not listening. Seconds that nothing was heard in are left as gaps in the chart rather than filled with zero, and the summary counts only the time the page actually sampled. If those two spans differ, the Duration figure says so underneath. A record that quietly claimed eight hours it did not hear would be worse than one that admits six.

The run stops on its own after twenty-four hours.

Nothing is recorded and nothing is uploaded

No audio is captured, buffered or sent. The microphone stream is turned into one number per second in the browser and thrown away frame by frame; there is no server in the path at all, because there is nothing to send.

The night itself — those one-per-second numbers, and the settings they were taken under — is written to this browser's own storage every fifteen seconds and whenever you switch away from the tab, so a reload or a crash does not cost you the night. It stays on the device. In the morning the page offers it back read-only, with a Clear this log button that removes it for good. If your browser refuses to store anything, which private windows do, the page says so and keeps recording regardless.

Taking it away with you

CSV is one row per second, with the timestamp in full and the level beside it. The file opens with a few commented lines naming the curve, the response, the correction and when it started, because a column of decibels is unreadable six months later without them. Seconds that were not heard keep their row and leave the value empty.

PNG is the whole night as one picture: the four figures, the chart with your marks drawn across it and the loud stretches shaded, and the line about what the numbers are. That last part is in the image on purpose. Pictures get forwarded and cropped, and the disclaimer should have to be cut out deliberately rather than left behind by accident.

What the file cannot do

A file of numbers is evidence of a kind, and it is not the whole case. What makes a noise record credible is that it was kept consistently, from the same place, over enough nights to show a pattern, with a line at the top saying what the figures are and are not — and with your own words about what you could actually hear, which no microphone records. How to build one is set out in document neighbor noise; this page is the part of that job that can be automated, and the rest of it still wants a person.

If what you need is a reading now rather than a record of a night, the online decibel meter on the front page is the same estimate with the session figures and the dose beside it.