What the research says
The most careful public tests of phone-based sound measurement come from the U.S. National Institute for Occupational Safety and Health (NIOSH).
- In a 2014 study, Kardous and Shaw tested 10 iOS apps on four Apple devices in a reverberant chamber, using pink noise from 65 to 95 dB and a laboratory reference microphone. Apps ran with their default calibration. The size of the average error per app ranged from 0.07 dB to more than 12 dB for unweighted levels, and from about 0.5 to 13 dBA for A-weighted levels. NIOSH reports that four of the ten apps were within ±2 dB of the reference on average.
- Results also varied by device. The same apps averaged closer to the reference on some iPhone models than on others.
- In a 2016 follow-up, the four best apps were retested with calibrated external microphones. Individual differences from the reference shrank from a range of −14 to +11.3 dB with built-in microphones to −1.4 to +1.8 dB. The authors concluded that the built-in microphone was the main source of error.
Two cautions when applying this to our meter. Those tests covered native iPhone apps, steady noise and a mid-level range. A browser meter has less control over the microphone than a native app, and NIOSH did not consider Android hardware consistent enough to verify. So we do not claim a fixed accuracy figure for an uncalibrated device.
Where the errors come from, and what we do about them
| Error source | What happens | What our meter does |
|---|---|---|
| Microphone sensitivity | Each device model turns the same sound into a different signal level, so the raw number is shifted up or down. | A calibration offset that you set once per device and browser. |
| Frequency response | Phone and laptop microphones are tuned for speech. They typically lose sensitivity at low bass and very high frequencies. | A, C and Z weighting. A-weighting gives less weight to the frequencies where small microphones are weakest, so dBA readings are usually the most trustworthy. |
| Operating system and browser processing | Echo cancellation, noise suppression and automatic gain control are designed for calls. They change the level, sometimes pushing steady noise down over time. | We ask the browser to switch all three off, then check what it reports. If processing could not be disabled, the meter shows a notice. |
| Clipping at high levels | Loud sound overloads the microphone or converter. The signal flattens and the reading stops rising. | A clipping badge appears when samples hit full scale. The displayed value is then a lower bound. |
| Noise floor | The microphone's own electronic hiss sets a minimum reading, so a very quiet room reads higher than it is. | Nothing can remove this in software. Treat very low readings as "at or below" the value shown. |
| Case, hand and position | A case or finger over the microphone port, wind, handling noise and reflections from nearby surfaces all change the result. | Hold the device away from your body with the microphone uncovered and pointed toward the source. Use Slow response and Leq for a steadier value. |
iPhone and iPad browsers deserve a special mention. Reports in the WebKit bug tracker describe cases where Safari kept processing or filtering the microphone signal even when a page asked it not to. That is why we check the reported settings and tell you, instead of assuming the request worked.
Our own device tests
We are comparing phones and laptops against a reference sound level meter and will publish the results in this table.
| Device | Browser | Reference meter reading | Our reading (uncalibrated) | Difference |
|---|
Device tests in progress — results will be published here.
What you can rely on
- Relative changes. On one device, "5 dB louder than before" is far more reliable than the absolute number.
- Steady, mid-level sound after calibration: traffic, appliances, office and classroom noise.
- Less reliable: very quiet rooms, very loud events, bass-heavy sound and short impulses such as claps or bangs.
For the exact calculations behind each number, read the methodology. For native apps, see the decibel meter app comparison.
Sources
- Kardous CA, Shaw PB. Evaluation of smartphone sound measurement applications. J Acoust Soc Am. 2014;135(4):EL186–EL192
- Kardous CA, Shaw PB. Evaluation of smartphone sound measurement applications (apps) using external microphones: a follow-up study. J Acoust Soc Am. 2016;140(4):EL327
- CDC / NIOSH: NIOSH Sound Level Meter App (accuracy and Android statement)
- WebKit Bugzilla 192932: Can not disable audio processing (lowpass)
- MDN: MediaTrackConstraints (echoCancellation, noiseSuppression, autoGainControl)
Frequently asked questions
How accurate is a phone as a decibel meter?
In NIOSH laboratory tests the best iPhone apps averaged within about 2 dB of a reference meter, while others were off by more than 10 dB. Accuracy depends on the phone, the software and whether it has been calibrated.
Why do two phones show different decibel readings?
Microphones differ in sensitivity and frequency response, and phones apply different audio processing. Two uncalibrated devices side by side can easily disagree by several decibels.
Can I use this meter for workplace or legal measurements?
No. Readings are estimates and are not suitable for occupational compliance, legal evidence or medical decisions. Those need a calibrated sound level meter.
What does the clipping badge mean?
The sound was loud enough to overload the microphone or its converter. The true level is at least as high as the reading shown, and probably higher.
How can I make the readings more accurate?
Calibrate against a handheld sound level meter, choose A-weighting when comparing with dBA values, hold the device with the microphone uncovered and avoid wind and handling noise.
Updated