Browser Decibel-Meter Accuracy and Limitations

What the number means

The panel shows an uncalibrated browser estimate derived from Web Audio time-domain samples. Gain, placement, case resonance and frequency response differ per device, so two phones side by side can disagree by many decibels. Readings are never presented as approved, professional or lab-tested instrument results.

Why results differ between devices and browsers

• Hardware variation: microphone capsules, preamp gain and placement differ across phones, laptops and headsets. Android devices vary especially widely.
• Operating-system processing: the OS may apply gain or filtering before the browser ever sees samples.
• Browser audio processing: echo cancellation, noise suppression and automatic gain control (AGC) reshape levels. This site requests them off via getUserMedia media constraints, but the browser or device may keep them enabled — the meter reports when that happens.
• Mobile versus desktop: phones use aggressive voice processing and sit close to hands and pockets; laptops have noisier fans and more distant mics.

What calibration improves — and what it cannot fix

A calibration offset corrects overall gain against a reference you trust. It does not correct frequency response, distortion, clipping or processing that changes with level. A profile is device-specific: change the microphone, position, weighting or response and the old offset stops applying. See how to calibrate a browser decibel meter.

When a certified meter is required

Use a calibrated Class 1 or Class 2 sound level meter — not this browser estimate — for workplace noise assessment, legal disputes, hearing-safety verdicts, or any claim of compliance with IEC, ANSI, OSHA or NIOSH requirements. This site shows no safety verdicts for exactly this reason. Background on meter practice: CDC/NIOSH noise and sound guidance.

What improves a casual reading

• Measure in a stable position at ear height, away from wind, vents and handling noise.
• Use A-weighting for speech-like environments, C or Z when comparing spectra (dB vs dBA).
• Let Fast/Slow smoothing settle for a few seconds before comparing with a decibel comparison chart.

Same kitchen, two phones: what a 5–10 dB spread looks like

Set two phones on a kitchen table, one metre from a running extractor fan, and press Start on both. You might see 62 dBA on one and 55 dBA on the other before any calibration. Neither is broken. One capsule sits behind a tiny hole near the charging port, the other sits beside the camera, and each maker applies its own preamp gain. Across common models that spread often lands around 5–10 dB, wider on Android because the hardware pool is so mixed. Your own check is simple: run the same 30–60 second average on each device, note the Leq and duration, and write both numbers down. The gap between them is the reason this site labels every uncalibrated reading an estimate and keeps raw dBFS visible in Digital Diagnostics.

Floor, ceiling and bass: where the estimate bends

A quiet bedroom at night might sit near 28 dBA on a reference instrument while your laptop shows 38. That high reading below roughly 30 dBA is the microphone self-noise floor talking, not your room getting louder. At the other end, a blender measured close up can flatten near 95–100 dBA and stop climbing even as the sound clearly gets harsher — the capsule is clipping. Bass is the third bend: a washing machine rumble or distant traffic thump often reads several decibels low because small capsules roll off low frequencies and the correction from the dominant band cannot restore what was never captured. When you compare against the decibel comparison chart, let Slow response settle, average a full half minute or longer, and round to whole numbers. Decimals imply a precision the hardware cannot back up.

Habits that quietly wreck a casual reading

• Reading the first jumping digits instead of the settled Leq — wait for Fast 125 ms or Slow 1 s to stabilise, then take current, minimum, Leq and maximum together with duration.
• Holding the phone while it measures — handling thumps and a covered port can swing the display 6 dB or more. Set it down, mic uncovered, at ear height.
• Measuring in wind, near vents or with notifications firing — one chime inside a 60-second average lifts the whole Leq.
• Comparing an instant peak to a chart built from averages — a door slam hitting 88 for half a second is not the same as 88 sustained.
• Treating idle dashes as silence — when the meter shows "--" it means no live input, never a simulated zero. Start the session over HTTPS in a current Chrome, Edge, Firefox or Safari build and grant permission once per origin before judging a low room.

Sources and technical references

The practical limits above are consistent with CDC/NIOSH guidance on sound-level apps and the browser permission model documented by MDN getUserMedia().

Next: calibration → · Methodology → · Open the online decibel meter →