The ACX noise floor: what −60 dB RMS actually asks of your room

dBFS, not dB SPL: the scale your file lives on

The dB values in audio files are dBFS — decibels relative to full scale — where 0 is the loudest value a digital file can hold and everything real is negative. This is not the dB of noise-meter phone apps, which measure sound pressure in a room (dB SPL). A phone app telling you your room is “30 dB” tells you nothing about whether your files will pass, because your floor in dBFS depends on the room and your mic's sensitivity and your gain setting, combined. The only measurement that counts is the one taken through your own chain at working gain — which is why the measurement guide starts there and never mentions phone apps again.

Why RMS, not peak

RMS (root mean square) is an average-energy measurement over time; peak is the single loudest sample. ACX specifies the floor in RMS because steady noise is what fatigues a listener over a ten-hour audiobook — a continuous hiss at a modest level is more damaging than one tiny tick, and an average captures the continuous character of the noise. Practically, this cuts both ways: a lone distant thump might barely move an RMS reading (good news), but no amount of “it's mostly quiet” rescues a floor with constant hiss running through it (the bad news, and the common case).

The standard measuring convention, used by tools like Audacity's free ACX Check plugin, is to read the RMS of the quietest half-second or so it can find in your selection — your truest available silence. That's worth remembering: the meter hunts for your best moment, and even your best moment has to pass.

What real rooms measure

Numbers from the field, so you know where you likely stand before you measure. These are typical ranges for a condenser mic at sane gain; your chain will vary, which is the whole reason to measure rather than assume:

1. Living room, daytime−38 to −48
2. Bedroom, evening−48 to −56
3. Same bedroom, late night−56 to −64
4. Closet full of clothes, quiet hour−60 to −72

−90−60 line−30 dB RMS

Bars are the typical ranges from the table; sage bars reach the passing side of the line.

Two lessons hide in that table. First, the spread between rows is mostly behavior — time of day, appliance discipline, mic distance, gain — not purchased acoustics. Second, the distance from “typical untreated room” to “passing” is usually 5–15 dB, which is exactly the range a light software chain can close cleanly. It's the 20+ dB gaps that force upstream changes, and diagnosis tells you which kind of gap you have.

−90−60 ← the line−30

Passing is necessary. It isn't sufficient.

The floor is a threshold, not a grade — and clearing it doesn't make audio sound clean. Three things the −60 number cannot hear: intermittent noises (clicks, thumps, chair creaks) that barely dent an average; the character of your floor (a smooth distant hush reads as professional at −60, while a whistling electronic hiss at the same −60 draws attention); and everything ACX's “free of extra sounds” clause covers, which is judged by ears, not meters. The reverse failure exists too: a floor gated to absolute digital silence measures gorgeously and sounds wrong, because human QA expects a living room tone under human speech. Natural-but-quiet beats dead-but-perfect — here's that story in full.

The three levers that move the number

Everything that lowers a noise floor is one of three moves, and they stack:

• Lower the noise reaching the mic — scheduling, switching things off, pointing the mic's rejection at the problem. Free, and where the biggest wins live. (The physical-treatment version of this lever is real too; soft mass beats software, and that single sentence is our entire coverage of it.)
• Raise the voice relative to the noise — get closer to the mic, keep performance energy up, stage gain so you're not amplifying the room to reach a healthy level. This is gain staging, and it's routinely worth 6–12 dB.
• Subtract what remains, in software — high-pass filtering plus gentle broadband reduction as the final polish. The chain guide covers how much is safe; the short answer is “less than the slider allows.”

Run in that order — it's the fix order in miniature — each lever leaves less for the next one to do, and the software lever stays in its polite range where artifacts don't happen.

Floor questions, answered straight

Is a lower floor always better?

Down to a point. Below roughly −70 dB RMS of natural room tone you're in genuinely professional territory; “floors” near digital silence usually mean over-processing rather than an extraordinary room, and QA ears notice the difference. Aim for comfortably under −60 with tone that still sounds like air, not for the biggest negative number.

My floor passes at night but fails at noon. Which is my real floor?

Both — a floor is a property of a moment, not a room. The professional move is to know your schedule and book your recording hours accordingly; the 24-hour floor protocol shows you how to map the swing in your own house.

Does the floor spec apply to auditions too?

Formally no — nobody meters an audition. Practically yes: casting ears do instantly what the meter does slowly. A floor that would fail ACX is audible in the first seconds of a self-tape, which is why audition audio borrows the same target.