Browser Sound Level Estimator

Uncalibrated. When you press START, the browser asks for permission to use your microphone and this page analyses the incoming audio. It shows the A-weighted level of that digital signal in dBFS — decibels relative to digital full scale. This is a real, honestly-labelled quantity, and it is genuinely useful: it tells you whether closing the window helped, which of two rooms is quieter, or when the noise next door starts. What it cannot tell you is the absolute sound pressure level, and it cannot be compared with a reading from a different device.

Calibrated. Place this device next to a trusted sound level meter, let both measure the same steady sound, and type that meter's reading into the calibration panel. We store the difference between the two as a single offset and add it to every later reading, which converts the digital signal level into an estimated A-weighted sound pressure level in dBA. The status directly under the reading always tells you which mode you are in.

Why calibration is unavoidable. A digital audio signal carries no information about the sound pressure that produced it. The relationship between the two depends on the microphone's sensitivity, its position and casing, and any processing the operating system applies. No browser API reports microphone sensitivity, so the overall level relationship has to come from outside — from one comparison against an instrument that is calibrated. That comparison fixes the overall level offset for the conditions you calibrated in, which is the largest single error; it does not correct frequency response, directionality, hardware compression or clipping, or undocumented processing, so the result stays an estimate rather than a certified measurement. A working microphone and microphone permission are required for any of this; if permission is blocked we say so and link to microphone troubleshooting rather than showing invented numbers.

The signal path is deliberately plain, with no step that exists to make the number look convincing. Each frame of audio is reduced to its root-mean-square amplitude and converted to dBFS. The spectrum of the same frame is then used to compute the A-weighting correction from the standard A-weighting transfer function, normalised so that 1 kHz is exactly 0 dB, and that correction is applied to the broadband level. The result passes through exponential time weighting — 125 ms for FAST, 1 second for SLOW — applied to sound energy rather than to decibel values, which is the mathematically correct way to average levels. In parallel, a rolling 60-second energy average produces the equivalent level. Finally, and only if a valid calibration exists, a single stored offset is added.

Deliberately absent: C-weighting, standardised impulse detection, certified peak measurement, and anything described as LAeq. Where the browser reports that automatic gain control or noise suppression is active, the page says so and treats any stored calibration as void, because gain control changes the signal level independently of the sound. Where the browser refuses to report those settings — common on iPhone, iPad and Safari — we tell you that we cannot confirm it rather than assuming a clean path. See how it works and how accurate phone decibel meters are.

The exposure guidance elsewhere on this site is real: NIOSH recommends a limit of 85 dBA averaged over eight hours with a 3 dB exchange rate, and the OSHA permissible exposure limit is 90 dBA over eight hours with a 5 dB exchange rate. Those figures describe measurements made with calibrated, A-weighted instruments by people trained to use them.

This page never calculates an allowable or remaining exposure time from its own reading — not from an uncalibrated one, and not from a calibrated one either, because this is not a validated occupational sound level meter or dosimeter. The published criteria above are reference material, kept separate from your reading. A reading here must never be used to decide whether an environment is legally or occupationally safe, and never to conclude that hearing protection is unnecessary. It is not suitable for workplace compliance, legal evidence, hearing diagnosis, medical decisions or certified acoustic surveys. Treat a high reading as a reason to reduce exposure or to measure properly — never as clearance.

Use the same device for both sides of any comparison, and prefer the equivalent level over the instantaneous reading: sixty seconds of energy average characterises a room far better than a snapshot. Compare at similar times of day, because urban noise commonly falls 10–15 dB between late afternoon and the small hours. Shield the microphone from wind outdoors, do not cover it, and hold the device away from your body so reflections and handling noise do not dominate. Do not try to measure your own device's speakers: the microphone sits close to them and vibration couples straight into the housing.

The single most important habit is to read the mode label before the number. In uncalibrated mode a difference of 6 dB between two rooms is meaningful; the value −38.4 dBFS on its own is not a statement about loudness at all. Browse the decibel level chart for published reference levels of everyday sounds.