The short answer
A phone can show you how a sound level changes, and roughly how loud a place is. Whether the number itself is right depends on four things: the phone’s microphone, the audio processing the phone applies, the math the app uses, and whether the app was calibrated against a real meter. Without that calibration, treat any phone reading as an estimate.
We make Sound Meter, a free decibel app for Android, so we tested this ourselves. Here is what we measured, what changed when we fixed our own app’s math, and what we still cannot prove without a reference instrument. The table is in What we measured on our own test phone.
What a real sound level meter has to do
The international standard for sound level meters is IEC 61672-1. It sets performance specifications for meters that measure time-weighted, frequency-weighted sound levels, time-averaged levels and sound exposure, and it defines two performance categories, class 1 and class 2. The acceptance limits for class 2 are equal to or wider than those for class 1 [1].
Two terms from that world explain most of what a decibel app shows:
- Frequency weighting. A-weighting turns down low and very high frequencies, roughly the way human hearing does, and gives readings in dBA. C-weighting is much flatter, and Z means no weighting at all. In the standard’s table, A-weighting is −19.1 dB at 100 Hz, −10.9 dB at 200 Hz and 0 dB at 1,000 Hz, while C-weighting is only −0.3 dB at 100 Hz [2].
- Time weighting and averaging. “Fast” and “Slow” smooth the reading with a time constant of 125 milliseconds or one second. An equivalent level, written Leq or LAeq, is the steady level that would carry the same sound energy over the whole measurement. Health limits are written in averaged levels like this, not in single glances at a number.
What NIOSH found when it tested apps
The US National Institute for Occupational Safety and Health (NIOSH) tested sound apps in its acoustics laboratory, using pink noise from 65 to 95 dB in a reverberant chamber, next to a reference microphone and a type 1 sound level meter. It examined 130 iOS apps and found 10 that met its selection criteria. Of 62 Android apps, only 4 partially met the criteria, so a full comparison of Android apps was not possible. For A-weighted levels, three apps came within ±2 dBA of the reference on average [3].
NIOSH put the ±2 dBA figure in context: national standards and occupational guidelines specify that accuracy for type 2 instruments. Its follow-up study in 2016 found that apps used with external, calibrated microphones measured within ±1 dB of the reference [3].
NIOSH’s own Sound Level Meter app is available only for iOS. NIOSH explains that the Android market is split among many manufacturers with different microphones and audio chips, so it cannot verify an Android app in its laboratory [4]. That is the honest position for any Android sound app, ours included.
What we measured on our own test phone
Before version 1.3.0, Sound Meter used a shortcut: it took the peaks of each block of audio and subtracted a fixed 2 dB to stand in for A-weighting. In version 1.3.0 (October 2026) we replaced that with an energy-based (RMS) level, real A and C filters and Fast and Slow time weighting. Before we released it, we ran a debug build that calculated both methods on exactly the same audio.
Method. Our test phone was a TECNO KJ5 running Android 13. On 1 October 2026 we played sounds from a MacBook speaker about 1 meter away at 50% volume. The first row is the room itself, with a fan running and nothing playing. Each row is the median of the per-block readings, A-weighted with Fast response. The values are uncalibrated: they show the difference between the methods, not the true sound pressure level.
| Sound | Old method (peaks, −2 dB) | New method (RMS, A-weighted) | Difference | New method, unweighted (Z) |
|---|---|---|---|---|
| Room with a fan | 42.4 | 32.2 | 10.2 dB | 33.5 |
| Recorded speech, 23 s | 68.8 | 55.1 | 13.7 dB | 62.6 |
| Steady 1 kHz tone | 57.4 | 54.0 | 3.4 dB | 54.0 |
The gap between the methods was not constant: 3.4 dB on a steady tone, 10.2 dB on fan noise and 13.7 dB on speech. A single calibration offset cannot fix an error that changes with the kind of sound. This is one reason two apps can show numbers 10 dB apart in the same room.
Worked example · why A-weighting lowered the speech reading
- The unweighted (Z) speech reading was 62.6. The A-weighted reading of the same audio was 55.1.
- 62.6 − 55.1 = 7.5 dB taken off by the A filter.
- The test voice was a low male text-to-speech voice. We attribute most of the 7.5 dB to its low pitch: A-weighting cuts 19.1 dB at 100 Hz and 10.9 dB at 200 Hz [2].
On the 1 kHz tone, where A-weighting is 0 dB, the A and Z readings were identical: 54.0 and 54.0. That is what a correct filter should do.
What we could not measure: we had no calibrated reference meter, so we cannot tell you how close any of these numbers is to the true level in that room. That is why Sound Meter marks its readings “approx.” and offers a calibration offset of up to ±20 dB.
A second test: a web page on an iPhone
On 30 September 2026 we tested whether a browser-based decibel meter could work on an iPhone. Our test page asked the browser to switch off echo cancellation, noise suppression and automatic gain control. On an iPhone running iOS 18.7, Safari turned off echo cancellation only; the other two settings were not offered at all.
We played loud, steady pink noise from the MacBook at full volume. The iPhone’s unweighted reading rose to 48.9, then fell to about 30 within the next three seconds, although the sound had not changed. When we replayed the same sound to the MacBook’s own microphone in Chrome, with all processing off, the reading on that step stayed within about 1.3 dB.
So the iPhone’s audio processing was turning the loud sound down. This test covers a web page in Safari on one iOS version, not native iPhone apps. It is still a useful warning: if a reading sinks while a noise stays steady, the phone may be adjusting the microphone level.
Four things between the sound and the number
- The microphone and its ceiling. Each phone model has its own microphone and audio chip, as NIOSH points out [4]. A very loud sound can also hit the top of the microphone’s digital range. Sound Meter shows “mic limit” when the audio reaches 98% of full scale, because the true level is then at least the number shown.
- The phone’s own processing. Gain control and noise suppression change levels, as the iPhone test above showed. On Android, Sound Meter records from the system’s default microphone source. Android also defines an “unprocessed” source, but only phones that support it deliver raw sound; on the others it behaves like the default [6]. We have not checked every phone.
- The app’s math. Peaks or energy, a real A filter or a fixed offset, and how the reading is averaged. Our table shows that these choices alone moved readings by 3 to 14 dB.
- Calibration. An offset learned against a reference meter, with a steady sound, corrects the phone’s overall sensitivity. It cannot correct a method that is wrong for some sounds and right for others.
When a phone reading is enough
A phone is useful for comparing: before and after a fix, one room against another, a machine on and off. It is also a reasonable first warning. NIOSH’s recommended exposure limit at work is 85 dBA averaged over 8 hours, with a 3 dB exchange rate: for every 3 dB increase, the allowable exposure time is cut in half [5].
| Average level | Time to reach 100% |
|---|---|
| 85 dBA | 8 hours |
| 88 dBA | 4 hours |
| 91 dBA | 2 hours |
| 94 dBA | 60 minutes |
| 97 dBA | 30 minutes |
| 100 dBA | 15 minutes |
If a phone shows numbers near these levels for long periods, take it seriously: NIOSH calls protecting your hearing at 85 dBA or more a good health practice anywhere, not only at work [5]. But for workplace compliance, a noise complaint, a court case or a building report, use a class 1 or class 2 instrument and follow the method the authority asks for. A screening app, including ours, is not evidence of a precise level.
Getting a more useful reading
- Measure over time. Use a timed average (LAeq) of 30 to 60 seconds instead of reading the number at one moment.
- Keep the setup the same. Same distance, same phone position, case on or off every time. Do not cover the microphone with your hand.
- Keep the screen on. On our test phone, Android sent the app silence once the screen turned off.
- Calibrate once, with a steady sound. If you can borrow a calibrated meter, match your app to it with a steady noise, not speech or music.
Questions people ask
Can a phone app replace a sound level meter for a noise complaint?
Usually not. A phone log can show when the noise happens. A formal complaint normally needs a calibrated instrument and the authority’s own method.
Why do two decibel apps show different numbers in the same room?
They may measure different things. In our own test, the old peak-based method read about 3 to 14 dB higher than an energy-based one on the same audio, depending on the sound. Weighting, averaging and the microphone add more differences.
Is dBA the same as dB?
No. dBA is a decibel level measured through the A-weighting filter, which turns down low frequencies. On the low voice in our test, the A-weighted number was 7.5 dB lower than the unweighted one.
Does calibrating an app fix its readings?
Only partly. Calibration against a reference meter corrects the phone’s overall sensitivity. It does not remove errors that change with the type of sound, the phone’s own processing, or the microphone’s limit.
Try it on your phone. Sound Meter shows A, C and Z weighting, Fast and Slow response and a timed LAeq, and marks its readings as estimates.
This guide explains how phone sound measurements work; it is not medical, legal or occupational safety advice. Ardesse is not affiliated with NIOSH, the CDC, the IEC or Google.
Sources
Last reviewed 2026-11-04. Our own measurements are from our internal test logs, named in each row of our fact-check file.
- IEC 61672-1:2013, Electroacoustics – Sound level meters – Part 1: Specifications (abstract) — read 4 Nov 2026.
- IEC 61672-1:2002, Table 2, frequency weightings (as reproduced in the iTeh Standards catalog) — checked 30 Sep 2026.
- NIOSH Science Blog: “So How Accurate Are These Smartphone Sound Measurement Apps?” (2014, with the 2016 update) — read 4 Nov 2026.
- NIOSH: Sound Level Meter App, technical specifications and FAQ — read 4 Nov 2026.
- NIOSH Science Blog: “Understanding Noise Exposure Limits: Occupational vs. General Environmental Noise” — read 4 Nov 2026.
- Android Developers: MediaRecorder.AudioSource (DEFAULT and UNPROCESSED) — read 4 Nov 2026.