Validation Protocol for Browser Sound Measurements

Required equipment

• Device under test (phone, tablet or laptop) with a current Chrome, Edge, Firefox or Safari over HTTPS.
• Reference instrument: calibrated Class 1 or Class 2 sound level meter, or an acoustic calibrator with a documented level.
• Stable sound source (loudspeaker playing pink noise or a steady tone) — not a phone speaker at full volume.
• Tripod or stable surface, tape measure, and a quiet room with doors closed.

Stable-source guidance

Warm up the source for at least 5 minutes. Choose a level near 70–80 dBA at the test position. Keep the room quiet: close windows, silence notifications, and avoid wind or direct airflow over either microphone.

Device placement

• Place the browser device and the reference microphone side by side, same height, same distance (e.g. 1 m from the source).
• Keep the browser microphone uncovered and pointed the same way for every run.
• Record microphone label, browser, OS and app version for every run.

Three repeated 60-second runs

1. Use A weighting and Slow response for all runs.
2. Start the browser meter, wait out stabilization, record 60 s without touching either device.
3. Note the browser LAeq, the reference LAeq, duration, sample rate, channels and processing flags.
4. Repeat twice more without moving anything. Compute the spread (max − min) of the three browser LAeq values.

A separate verification level

Change the level by at least 10 dB (or halve the distance), then capture one more 60-second browser average plus the reference average. Apply the calibration offset from the first level and report the residual (browser estimate − reference). A residual within 2 dB is consistent at that point; 2–5 dB means use with caution; over 5 dB means the calibration did not track consistently. These bands are product guidance, not certification.

Example logbook entry with real numbers

A usable entry fits in one paragraph. Test phone, mid-range handset on tripod, Chrome over HTTPS, mono input, processing flags reporting gain control off. Pink noise at 1 m, room 4 by 5 m, doors closed, background 34 dBA. Three 60-second runs on A weighting with Slow response give browser LAeq values of 73.8, 74.5 and 73.9 against a reference holding 74.2, 74.3 and 74.1. Spread is 0.7 dB, mean residual about −0.2 dB. Move the speaker to 0.5 m for the separate verification level: reference 84.1, browser 83.4 after applying the first-level offset, residual −0.7 dB. Sample rate, channel count and duration ride along on every line. Anyone reading that entry can rerun the geometry tomorrow and expect the same shape, which is the entire point of writing it down this way.

How to read spread against residual

Spread answers whether the browser repeats itself: take max minus min across the three LAeq values at one level. Under about 1 dB is tight for consumer hardware; past 2 dB, look for gusts, handling or an unstable source before blaming the meter. Residual answers whether the calibrated estimate tracks the reference: browser estimate minus reference after the offset is applied. Within 2 dB at the verification level counts as consistent for casual use, 2–5 dB means proceed with caution and shorter claims, and over 5 dB means the offset did not transfer — often frequency shaping, compression that shifts with level, or clipping near 95–100 dBA. Report both figures side by side. A meter can repeat within 1 dB yet sit far from the reference when uncalibrated, and a single offset never repairs bass roll-off, distortion or processing that the browser refused to disable. Cross-check the chain in how the meter works when the two disagree.

Runs that should never reach the logbook

• Music or speech as the source — syllables swing the average several decibels between runs; use steady noise or a tone.
• A phone speaker at full volume as the source — distortion and limiting make the reference itself unstable.
• Hand-held devices, moved tripods or a mic brushed mid-run — touch alone can outweigh the effect you hoped to validate.
• Mixed settings across repeats — keep A weighting and Slow response fixed, and note EC/NS/AGC state every time.
• Publishing only the closest run — log all three repeats plus the separate level, with duration and background level, even when one line looks untidy.

Required details to publish

• Device model, OS version, browser version, app version.
• Microphone label, sample rate, channel count, EC/NS/AGC state.
• Reference instrument make, model, calibration due date, weighting and response used.
• Source type, distance, room description and background level.

Suggested data table

Run | Browser LAeq | Reference LAeq | Residual | Duration | Sample rate | Processing | Notes. Publish all three repeats plus the separate verification level; never publish only the best run.

Repeatability versus absolute accuracy

Repeatability is how closely repeated runs agree with each other. Absolute accuracy is how closely the browser average agrees with the reference. A browser meter can repeat within 1 dB yet sit 10 dB from the reference when uncalibrated — both numbers must be reported separately. A single offset cannot fix nonlinearity, frequency response, AGC or clipping.

Limitations of browser and consumer microphones

Consumer microphones add unknown gain, frequency shaping, compression and browser processing (echo cancellation, noise suppression, automatic gain control). They are not Class 1 or Class 2 instruments. Browser readings are reference-only screening estimates — never use them for occupational, legal, medical or regulatory decisions.

Verified results

Open the online decibel meter → · Methodology → · Accuracy →