How to measure your noise floor

What you're measuring, precisely

Your noise floor is not your room's loudness — it's the level your whole chain prints when you're silent: room ambience plus mic self-noise plus preamp hiss plus every electrical gremlin between capsule and file, all scaled by your gain setting. Change any link — different gain, different mic distance for the voice that will sit above it, computer moved a meter — and the number changes. That's why a measurement is only valid taken through your own setup, at the settings you actually record with, at an hour you actually record. Phone-app dB readings measure the air, not your chain, and can't tell you anything about passing ACX's −60 dB RMS floor spec.

The protocol: silence, recorded honestly

1. Set up exactly as if recording

   Same mic position, same interface gain you use for real takes (if you haven't set that deliberately, do gain staging first — an arbitrary gain makes an arbitrary measurement), same software, same chair, door closed, you in position. The measurement must capture the session, not an idealized version of it.

2. Record 20 seconds of silence

   Hit record and hold still: no talking, no swallowing theatrics, no chair shifting — but breathe normally, because you'll be breathing during real takes. Twenty seconds gives analyzers plenty of clean material to hunt through. If a one-off noise intrudes (a car, a door), let the recording run longer rather than restarting; the analysis finds the quietest stretch.

3. Read the RMS

   In Audacity, the free ACX Check plugin (a community-built analyzer listed in Audacity's own plugin directory; installs into the Analyze menu) reads three numbers at once — peak, RMS, and noise floor — with the floor taken from the quietest half-second it finds in your selection. Select your silence, run it, read the floor line. Any DAW with RMS statistics works too: select the silence, open amplitude/loudness statistics, read average RMS. What doesn't work is the recording meter's bouncing peak display — peaks and RMS are different animals, and the spec is written in RMS.

4. Write it down, with context

   The number alone is half the data. Log the date, hour, gain setting, and what was running in the house. Three measurements at different hours tell you more than one perfect one — floors move with the household's schedule, sometimes by double-digit dB across a single day.

Reading your number

1. Above −40Chain problem
2. −40 to −50Loud problem
3. −50 to −60The workable gap
4. ACX line, −60
5. −60 to −70Passing
6. Below −70Professional territory

Every band and name comes from the table below, in dB RMS at working gain.

Measure at these moments, forever

A floor measurement is a snapshot, so take snapshots when the picture might have changed: before any big project (an audiobook contract deserves a fresh number, not last spring's), after any equipment or settings change, when seasons change the household (heating and cooling are the loudest roommates most narrators have), and whenever a client or QA flags noise. Narrators who log floors over time also get an early-warning system for free — a floor that quietly drifted 6 dB up since June is a fact you can investigate, instead of a vague feeling that recent files sound worse.

The two classic measurement mistakes

Measuring at the wrong gain. A floor measured at half your working gain flatters you by exactly the difference; first-time measurers do this constantly by turning the knob down “since I'm not talking anyway.” The measurement inherits meaning from your real settings — freeze them first.

Measuring processed silence. If your recording path already has suppression or reduction enabled — an interface's built-in processing, an OS “voice enhancement,” a suppressor app in the chain — you're measuring the software, not your setup. Measure raw, so you know what you actually have; then measure again after your chosen chain, so you know what it delivers. The difference between those two numbers is your software's real contribution, and knowing it keeps you from over-trusting either half. Where each kind of processing belongs is mapped in the software chain guide, and the whole method lives inside the fix order.

Measurement questions

Why 20 seconds and not five?

Because analyzers hunt for a clean stretch, and short recordings can be all contamination — one distant car and a five-second sample is useless. Twenty seconds nearly always contains the honest quiet moment the measurement needs.

My ACX Check floor and my DAW's RMS reading disagree. Which is right?

Both, about different things: ACX Check reports the quietest half-second, while a whole-selection RMS averages in every rustle. For spec purposes use the ACX-style reading; for honesty about your room, look at both.

Should I measure with noise reduction already applied?

Measure raw first, always. A raw floor is diagnostic — it tells you what your space and chain are doing. A processed floor only tells you what your plugin is hiding, and hiding is not the same as fixed.