How Accurate Are Phone Decibel Meters?

NIOSH screened 130 iOS sound-measurement apps against selection criteria, then tested ten of them plus four Android apps in a reverberant chamber against a reference system, using pink noise at 65 to 95 dB in 5 dB steps with the phone about six inches from the reference microphone.

The best iOS app differed from the reference by a mean of 0.07 dB unweighted and −0.52 dBA A-weighted. Three of the ten iOS apps tested came within ±2 dBA of the reference. The Android apps showed wide variance and did not meet the study's accuracy criteria — which the authors attribute to the diversity of Android hardware and audio paths rather than to the apps alone.

A follow-up NIOSH study tested calibrated external microphones — an iMM-6 and a MicW i436 — on six iPhones against reference instrumentation. With external microphones the mean difference was −0.023 dB with a standard deviation of 0.530 dB, and a total range of −1.4 to +1.8 dB.

With the phones' internal microphones, the same comparison gave a mean difference of 1.646 dB with a standard deviation of 3.795 dB, and a range from −14 to +11.3 dB.

That is the sharpest single result in this literature. The microphone, not the software, is the dominant error source: swapping it turned a ±14 dB problem into a ±2 dB one.

Murphy and King tested 100 smartphones and seven apps, taking 1,472 measurements in an ISO 3741 reverberation room. Overall, apps read +1.29 dB(A) high on average — with a standard deviation of 8.12 dB.

Broken down by platform: iOS averaged +2.93 dB(A) (SD 6.81, n = 1,052) and Android −2.79 dB(A) (SD 9.58, n = 420). Broken down by level, the error was strongly level-dependent: +5.33 dB(A) at a background level of 27 dBA, +2.09 at 50 dBA, +1.33 at 70 dBA and −3.57 at 90 dBA.

That level dependence is the practically useful finding. Phones overestimate quiet environments — because they are running into their own noise floor — and underestimate loud ones, because the microphone and audio path compress. The best-performing app in the study still came within ±2 dBA.

Pulling the three studies together, five things follow that apply to whichever app you install.

Every study above tested native applications. We searched for equivalent peer-reviewed evaluation of browser-based measurement through the Web Audio API and found none. So we cannot tell you how our own meter compares with a reference instrument, and we are not going to publish an accuracy claim we have not evidenced.

What we can state about the browser path specifically: the browser sits above the operating system's audio pipeline, which may apply processing we cannot see; automatic gain control is disabled where the browser allows it, but not every browser allows it; and a browser has no access to a device's microphone sensitivity, so there is no factory figure to work from.

This is precisely why our meter reports A-weighted digital signal level in dBFS by default, labelled as not being sound pressure, and only produces an estimated dBA figure after you calibrate it once against a meter you trust. The measured evidence on this page is the reason for that design, not a decoration on it.

Comparisons on one device are the strong case. Whatever offset your hardware has, it applies to both readings, so "is this room quieter than that one" and "did stepping back 10 metres help" are answerable.

Screening is a reasonable second case: if an uncalibrated app suggests you are somewhere near occupational thresholds, that is a signal to get a real measurement, not a measurement itself.

Compliance, legal evidence and hearing-conservation records are the cases where none of this suffices. Those need a Class 1 or Class 2 meter with a documented calibration.