Ask an acoustician how loud something is and you'll get a counter-question: at what frequency? A single decibel number is a summary that hides everything interesting. The neighbour's bass comes through your wall while their cymbals don't; a ventilation system drones at one pitch and hisses at another; the same 60 dB can be a tolerable rush or an unbearable hum. Professional acoustics therefore measures in frequency bands — slicing the spectrum into standardized chunks, classically octaves (each band twice the frequency of the last) or thirds of octaves, and reporting a level per chunk. There's an international standard just for the slicing: IEC 61260-1, which specifies the filters down to their tolerance masks.
Sentio's band-levels tool is the phone's version of that idea, and this post is about both halves of that sentence: how the phone genuinely reproduces the band-measurement structure, and precisely where it parts company with the calibrated instruments — because that boundary is sharp, and pretending otherwise is how measurement apps earn their reputation.
Sixteen tones, one honest trick
The tool doesn't passively filter ambient noise into bands. It's the receiving half of a two-phone measurement: a second phone (or the web app) plays a stepped tone sweep — sixteen pure tones, logarithmically spaced from 100 Hz to 12 kHz, roughly half an octave apart, each held for a beat with a gap before the next. The measuring phone listens through the raw, unprocessed microphone path and, for each tone, finds the spectral peak, measures its level, and measures the noise floor around it.
The elegance is that the sweep is self-identifying. Each tone announces itself by simply being the only tone playing, so the two phones need no cable, no clock sync, no pairing — the receiver recognizes each band by frequency and tolerates any phone's timing. And because every band records its own signal-to-noise ratio, the tool knows per band whether it truly heard the tone or just the room. A band that can't clear a 20 dB signal-to-noise bar is excluded from any conclusion rather than quietly diluting it — the same refuse-don't-guess rule the hearing screen applies to its room check.
This is the machinery underneath the two-phone sound insulation activity: sweep with a clear path, sweep again through the wall, and the per-band difference is the wall's attenuation spectrum — with both devices' unknown frequency responses cancelling in the subtraction.
Now, the calibration gap — stated plainly
A professional sound level meter is governed by IEC 61672-1, which defines two accuracy grades: class 1 (tighter tolerances, the lab and consultancy grade) and class 2 (slightly looser, the general field grade). The class isn't a marketing tier — it's a set of toleranced requirements on absolute sensitivity, frequency response, linearity across level, directional response and temperature behaviour, verified on the actual instrument, plus a field calibrator screwed onto the microphone before and after every serious measurement session.
A phone fails this in one specific, non-negotiable place: the absolute reference. The tool's levels are decibels relative to the phone's own digital full scale (dBFS) — an honest, precisely measured number whose relationship to actual sound pressure in the room is unknown, because no one ever measured this microphone against a calibrator. Two phones reading the same tone will report different dBFS levels; both are internally consistent; neither is a sound pressure level.
Could an app just add a correction table per phone model? Some try. The NIOSH evaluation of smartphone sound measurement apps found the best apps landed within about 2 dB of reference instrumentation — impressive, and still not a class 2 certificate, because a model-average correction can't know about your particular microphone, your case, your pocket lint, or the automatic gain control that some audio paths silently apply. Which is why this tool records which audio path it captured through and flags any capture where gain riding may have occurred: a level measurement on a gain-controlled path isn't a worse measurement, it's not a measurement.
The design answer: only differences leave the phone's frame
The band-levels tool resolves the gap the way the whole platform does — by building claims out of quantities where the unknown calibration cancels:
- A difference between two captures on the same phone (the insulation activity's reference-minus-through) subtracts the microphone's unknown response out of the answer entirely.
- A shape across bands — bass-heavy versus treble-heavy leakage — survives calibration error, since the error is fixed while the shape varies.
- An absolute dBFS level in one capture stays in the record for reproducibility, labelled as what it is, and is never promoted into a "the room is X decibels" claim.
| Professional instrument | Phone band-levels tool | Where the gap bites |
|---|---|---|
| Class 1 sound level meter + octave analysis ($3,000–$10,000, annual lab calibration, field calibrator) | 16-band tone-sweep levels in dBFS, per-band SNR, unprocessed audio path | Any absolute SPL claim: noise ordinances, occupational exposure, environmental assessment — certificate required, phone disqualified |
| Class 2 meter ($300–$2,000, IEC 61672 verified) | Same-phone level differences, in which device response cancels | Comparing your reading to anyone else's, or to a regulatory threshold — dBFS doesn't cross devices; differences do |
| Building-acoustics rig (calibrated source + meter, ISO field methods; typically a hired consultant) | Two-phone sweep: relative attenuation spectrum of your wall, your geometry | Official ratings (Rw/STC) for building certification follow standardized rooms and methods — the phone's number is yours, not comparable to a datasheet |
| IEC 61260 octave-band filters, toleranced and verified | FFT peak levels at 16 known tone frequencies — band sampling, not toleranced band filtering | Broadband and impulsive noise (traffic, machinery bursts) needs true filter banks; the sweep method only measures what its tones probe |
The honest claim, in one sentence: the phone measures the shape of sound and the change in sound with real fidelity, and does not measure sound pressure at all. Every result screen in the tool is worded from inside that boundary — "~28 dB median attenuation of this rig and geometry", never "your wall is rated 28 dB".
What that's still worth
A great deal, it turns out. The questions people actually have — does sealing the window gap help, which wall is the weak one, is the new door better than the old, why do I hear the neighbour's subwoofer but not their TV — are all difference-questions and shape-questions, squarely inside what the phone measures well. The consultant with the class 1 meter answers a question you'll ask at most once per building; the phone answers the twenty questions before and after it, including whether the fix you paid for did anything. Measured the day the work finished and filed under the same wall in your catalog, that's a before/after no invoice includes.
For the passive single-phone view of a sound's spectrum, see the frequency fingerprint; for the full argument about which measurements a phone should and shouldn't attempt, the honest accounting.