What dBFS Means on a Mic Level Meter

A decibel is a ratio, not an amount

The first thing to unlearn is that a decibel is a fixed quantity of loudness. It is not. A decibel expresses a ratio between two values on a logarithmic scale — it always answers "how many times bigger or smaller than the reference?", never "how much, absolutely?". A dB figure on its own is meaningless until you know what it is measured against, which is why the unit almost always carries a suffix: the letters after "dB" name the reference.

For amplitude — the size of a signal's swing, which is what an audio meter tracks — the formula is dB = 20 × log10(ratio). Two consequences are worth memorizing. Halving the amplitude changes the level by about −6 dB, and one tenth of the amplitude is exactly −20 dB. So a signal reading −20 dBFS is not "twenty percent quieter than maximum"; it is running at one tenth of the full-scale amplitude. The logarithm compresses an enormous range of signal sizes into small, readable numbers, which is the whole reason meters use it.

One footnote for the curious: for power quantities the multiplier is 10 rather than 20. This page converts an amplitude ratio, so 20 is the multiplier for the formula and table shown here.

Full scale: a nominal reference, with a floating-point exception

FS means full scale: the nominal maximum magnitude of a digital sample representation. In fixed-point PCM and at a constrained output, 0 dBFS is the reference normally associated with the largest representable sample magnitude, and values below that reference are negative. That model is useful—but Web Audio processes 32-bit floating-point samples. The MDN source below explicitly notes that analyser values have a nominal −1 to +1 range and can exceed it after operations such as down-mixing.

Device Bench makes its own display choice explicit. Its formatter shows 0.0 dBFS for readings at or above −0.05 dBFS and counts an inspected window as clipping when a sample reaches the engine's −0.1 dBFS threshold. That is evidence that this browser stream reached the meter's near-full-scale threshold. It is not proof of where upstream clipping occurred, and the page deliberately does not expose a positive floating-point internal value as extra unclipped headroom.

Average versus peak: why the meter shows two numbers

Open the microphone test and you will see two readouts: Level and Peak. Level is an RMS-derived figure — root mean square, an average of the signal's energy over each short analysis frame. It moves smoothly and tracks how strong your voice is overall. Peak holds the highest level the meter has seen since the test started, so it remembers the loudest moment even after the moment has passed.

You need both because speech is spiky. Plosive consonants and emphasized syllables push far above the average of the surrounding sound, so a voice averaging −20 dBFS can still throw peaks well above that. The average tells you where your voice sits; the held peak tells you how close your loudest instants came to the ceiling. Judging a mic level by average alone is how recordings that "looked fine on the meter" end up clipped on every hard consonant.

What dBFS is not

dBFS is routinely confused with dB SPL, and the two share almost nothing but the letters. dB SPL measures sound pressure in the air against a fixed physical reference — 20 micropascals, roughly the quietest sound a human ear can detect. It describes the acoustic world. dBFS describes only the digital one: how much of the format's numeric range a signal occupies. The same voice at the same distance can read −40 dBFS or clip at 0.0 depending on nothing but the input gain setting, so a dBFS number tells you nothing about how loud a sound actually was in the room.

It also is not perceived loudness. Human hearing weights some frequencies far more than others, and a plain amplitude meter ignores that entirely. Broadcast and streaming platforms measure program loudness in LUFS, a loudness-weighted cousin of dBFS — but that is a different meter for a different job, and it is not what a live input meter shows. dBFS answers exactly one question: how much of the digital range are you using?

How to read the microphone test

Start the test, grant mic access, and speak at your normal distance in your normal voice — reading a sentence aloud works better than saying "test" once. Watch the Level readout while you talk, then check where Peak has settled. Device Bench highlights −30 to −12 dBFS as a practical working band for this browser meter, not as a universal recording or broadcast standard. Different microphones, operating systems, processing modes, and applications can call for different gain. A level near the meter floor supports trying distance or input gain; a Peak displayed at 0.0 means this stream reached the meter's near-full-scale threshold, not that the page located upstream clipping.

The gap between a displayed peak and 0 dBFS is practical headroom on this capped readout: room for a laugh, cough, or raised voice before the meter reaches its threshold. Web Audio's floating-point graph can exceed the nominal full-scale magnitude internally, as described above; Device Bench deliberately does not display that excess as extra usable input headroom.

The useful mental model is a ratio scale, not a physical loudness verdict. Negative readings are below the nominal reference; this page's 0.0 display means its threshold was reached. Adjust one setting, repeat the same phrase and distance, and compare the result instead of inferring where limiting occurred.

A reference table from the formula

For an amplitude ratio a, the level is 20 × log10(a) dBFS. The table is calculated from that formula, not chosen as a set of descriptive bands. Halving amplitude is about −6.02 dB; dividing it by ten is exactly −20 dB.