How the Sound Decibel Meter Works

The Measurement Chain at a Glance

Each step is explained below with the settings the meter really uses. The numbers come from the tool’s own build notes and test report, not from a generic description of how decibel meters work.

Step 1: You Press Start and the Browser Asks for the Microphone

Nothing listens until you press Start. The page then calls the browser’s getUserMedia function, and the browser, not this site, shows the permission prompt. If you block it, the meter tells you how to allow the microphone in your browser settings, and the Start button stays usable so you can try again.

The meter asks the browser to switch off echo cancellation, noise suppression and automatic gain control, because all three change the level of what the microphone hears. Some phones and headsets ignore that request and process the audio anyway, which is one reason readings differ between devices. If you are not sure your microphone works at all, the microphone test shows its level, device details and whether that processing is on.

Measurement stops when you switch tabs or hide the page, so the microphone is never left running where you can’t see it. While you measure, the meter asks the device to keep the screen awake.

Step 2: Web Audio Delivers Raw Samples

The browser hands the page a stream of numbers between −1 and +1, usually 44,100 or 48,000 of them per second. They are not pressure values: they are the microphone’s output after the device’s converter, scaled so that ±1 is the loudest signal the device can record (0 dBFS, “full scale”). The page reads blocks of 2,048 samples and, on every screen refresh, uses only the samples that are new since the last one, so no sound is counted twice.

Step 3: A, C or Z Frequency Weighting

Human hearing is less sensitive to low and very high pitches, so sound level meters weight the spectrum. The meter builds its A- and C-weighting filters from the pole frequencies in the sound level meter standard, IEC 61672-1 (20.6 Hz, 107.7 Hz, 737.9 Hz and 12,194 Hz), converts them into digital filters for your device’s actual sample rate, and sets them to exactly 0 dB at 1 kHz. Z weighting is flat: no filter.

In testing, both filters stayed within ±0.3 dB of the standard’s table up to 4 kHz at 44.1, 48 and 96 kHz. A 100 Hz tone read 19.0 dB lower in dB(A) than in dB(Z), where the table gives 19.1 dB. dB(A) is the default because hearing-safety limits are written in dB(A); dB(C) suits bass-heavy sound such as music. The difference is explained in dBA vs dBC.

Step 4: From Samples to Energy in dBFS

For each block of new samples, the meter squares every value and averages them: the mean square, which is RMS squared. The level relative to full scale is 10 × log10(mean square). A full-scale sine wave comes out at about −3 dBFS, and every 10 dB step is ten times the energy. Why dBFS is a different unit from the dB on a sound level meter is covered in dB vs dBFS.

Step 5: Fast or Slow Time Weighting

A raw level jumps around too quickly to read, so the mean square is smoothed with an exponential time constant, as on a hardware meter: Fast uses 125 milliseconds and Slow uses 1 second, the two time weightings defined in IEC 61672-1. Fast follows speech and short events; Slow is easier to read for steady noise such as traffic or a fan.

Step 6: Offset and Calibration Turn dBFS into an Estimated dB Level

A browser never learns how much sound pressure a sample value stands for, because that depends on the microphone’s sensitivity and the device’s gain. The meter therefore adds a fixed nominal offset of +100 dB to the dBFS value. A tone captured at −23 dBFS is shown as about 77 dB; in testing, a 1 kHz tone at that level read 76.9 dB(A).

Your own calibration offset is added on top, anywhere from −30 to +30 dB in 0.5 dB steps. If you have a reference sound level meter, the Match a reference meter button compares the last 2 seconds with the value you type in and sets the offset for you. The offset is saved in your browser on that device only. The full procedure is in how to calibrate your microphone for a decibel meter, and the unit being estimated is explained in what sound pressure level means.

Step 7: The Numbers on the Screen

• Current: the time-weighted level right now.
• Minimum and maximum: the lowest and highest time-weighted levels of the session. The first half-second is skipped while the filters settle.
• Average (Leq): the energy average of the whole session, kept as running totals so memory stays flat for hours. Loud moments count for more than in a simple average; see what Leq means.
• Peak: the highest single sample of the weighted signal. It is not the C-weighted peak (LCpeak) that a professional meter reports.
• Session time, graph and files: a live graph of the last minute, the last 10 minutes or the whole session, a CSV file with one row per second, and a report image.

Next to the reading, the meter says what the level is roughly as loud as, highlights the matching row of an 11-step loudness scale and links to the matching “Is N dB loud?” page; the full list of everyday sounds is in the decibel chart. At 80 dB(A) and above it also shows the NIOSH recommended daily time, 8 hours ÷ 2^((L − 85) ÷ 3). A warning level (85 dB by default) turns the display red with a text label, not colour alone. If the input stays below −80 dBFS for 3 seconds, the meter warns that the microphone may be muted or covered; if the signal hits full scale, it warns that the input is clipping and the real level may be higher than shown.

Why Two Phones Show Different Numbers

The maths above is identical on every device. What differs is the hardware and software in front of it:

• Microphone sensitivity: the same sound produces different sample values on different models, and one fixed offset can’t fit them all.
• Built-in processing: some devices apply noise suppression or gain control even when asked not to. iPhones apply their own microphone processing, so calibration matters more there.
• Self-noise: a device microphone has its own hiss, often around 30 dB(A) or more, so a very quiet room reads higher than it really is.
• Clipping: very loud sound can overload the input, and the reading then shows less than the real level.
• Placement: distance, a case over the microphone hole, pockets, wind and handling noise.

That is why a browser estimate is not a Class 1 or Class 2 sound level meter reading, even after calibration. How far off a phone can be, and how to narrow the gap, is covered in how accurate an online decibel meter is.

Privacy: the Audio Stays on Your Device

All the processing above happens inside your browser. The tool scripts contain no code that sends audio or readings to a server. What stays on your device:

• Your calibration offset and the meter’s last 5 sessions, saved in the browser’s local storage. Clearing the site data removes them.
• CSV files and report images, which are created in the page and saved straight to your downloads.
• In the microphone test, a short recording you ask for: it is kept in the page’s memory for playback and download and is never uploaded.
• Audio files you open in the frequency analyzer or the loudness meter, which your browser decodes without sending them anywhere.

The site uses Google Analytics 4 for page statistics; it does not receive microphone audio. Details are in the privacy policy and on the data security page.

The Same Engine in the Other Tools

Every microphone tool on the site uses the same weighting filters, the same +100 dB offset and the same saved calibration, so a calibrated device reads alike across them. The calculators use published formulas instead of the microphone.

All of them are listed on the sound tools page.

What the Meter Can and Can’t Tell You

Use it to compare rooms, spot changes, check a noise against the loudness scale and learn how levels behave. Don’t use it as the only evidence for workplace compliance, a legal complaint or a hearing diagnosis: those need a calibrated Class 1 or Class 2 instrument and a qualified professional.

Each step described here was checked with reference signals before release. The results for every tool, including what could not be tested, are on the testing methodology page.