How our decibel meter measures sound

The signal chain, step by step

1. Capture. The browser opens your microphone with getUserMedia. We explicitly requestechoCancellation, noiseSuppression and autoGainControl to be switched off, because these voice-call features change the level of the signal and would make readings unreliable. Audio runs at the device’s native sample rate, usually 44.1 or 48 kHz.
2. Processing on the audio thread. Every sample is processed in an AudioWorklet, which runs on the browser’s real-time audio thread. Nothing is buffered to disk or sent over the network.
3. Frequency weighting. A and C weighting are built from the analogue pole frequencies defined in IEC 61672-1 (20.6 Hz, 107.7 Hz, 737.9 Hz and 12,194 Hz), converted to digital filters with a pre-warped bilinear transform and normalised to 0 dB at 1 kHz. Z weighting is flat, with a gentle 10 Hz high-pass that removes DC offset and handling rumble.
4. Time weighting. The squared, weighted signal is smoothed with an exponential average using the standard time constants: Fast (125 ms) or Slow (1 s). This is the value shown as the live reading.
5. Conversion to decibels. The level is 10 × log10(mean square) plus a device offset. The default offset maps digital full scale to roughly 100 dB SPL, a typical value for phone and laptop microphones. Calibration adjusts this offset for your specific microphone.
6. Statistics. Leq is the energy average of the weighted signal since you pressed Start, which is the figure most noise guidelines refer to. Max and Min track the time-weighted level after a short settling period, so the start-up transient does not distort them. Peak is the highest instantaneous value of the weighted signal.

Calibration

Every microphone has a different sensitivity, so an uncalibrated reading can be several decibels off. To calibrate, place your device next to a reference sound level meter (or use an acoustic calibrator at 94 dB), play a steady sound such as pink noise or a 1 kHz tone, and enter the difference in Settings. The offset is stored locally for that microphone and is applied automatically next time.

Accuracy and known limits

• Microphone frequency response. Built-in microphones are tuned for speech. Many roll off below about 100 Hz and some are uneven at high frequencies, which affects C and Z weighted readings more than A weighted ones.
• Noise floor. Phone and laptop microphones typically have self-noise around 20 to 35 dB(A). Readings in very quiet rooms may show the microphone’s own hiss rather than the room.
• Clipping at high levels. Depending on the device, the microphone or its converter saturates somewhere between about 100 and 120 dB. Above that, readings are a lower bound.
• Operating system processing. Some devices apply their own processing even when the browser asks for a raw signal. Calibration corrects a constant offset but cannot undo level-dependent processing.
• Not a certified instrument. A phone is not a Class 1 or Class 2 sound level meter under IEC 61672. For legal disputes, workplace compliance or regulatory reports, use a certified, calibrated meter.

In everyday ranges (roughly 35 to 100 dB(A)), an uncalibrated phone is usually within a few decibels of a reference meter for steady sounds, and a calibrated one can come considerably closer. Results vary by device, so if accuracy matters, compare against a reference and calibrate.

Check it yourself

You do not need to take our word for it. Normal conversation at one metre should read roughly 55 to 65 dB(A), and a quiet bedroom at night usually sits between 25 and 35 dB(A). If you have access to a reference meter, measure the same steady sound with both and compare the Leq values over 30 seconds.

Privacy

All analysis happens on your device. SoundsMeter never records, stores or uploads audio. See theprivacy policy for details, or read more about the project.