Home voice-over noise: the fix order

This page is the whole method, end to end. Every step links to a deeper guide, but you can run the entire loop from here in an afternoon. It assumes any working mic and any recording software — the method doesn't care which.

The number that decides everything

Your noise floor is what your recording chain prints when you say nothing: room ambience, electronics hiss, and whatever the house is doing, summed and measured in dB RMS. It is the single number that separates “sounds homemade” from “sounds professional,” because listeners forgive an imperfect voice long before they forgive a floor that hisses, hums, or breathes between phrases.

It's also the number the industry wrote down. ACX — the audiobook gateway feeding Audible and Amazon — requires every uploaded file to keep its floor below −60 dB RMS, with overall RMS between −23 and −18 dB and peaks under −3 dB. We re-verified those figures against ACX's published requirements on 2026-09-20; the full checklist lives in our ACX audio requirements guide. Even if you never narrate an audiobook, that trio is the most useful target in home VO: commercial and e-learning clients rarely publish a floor spec, but files that pass ACX pass everyone.

The trio from the verified note above, plus step 1’s typical untreated room (−45 to −58). Only the floor is a property of your room; RMS and peaks are set in mastering.

Why the order matters more than the tools

Noise problems stack multiplicatively, and software sits at the end of the stack. If your interface gain is 12 dB hotter than it needs to be, everything — room, hiss, fridge — arrives 12 dB hotter, and the plugin that has to remove it must work 12 dB harder, which is exactly when noise reduction starts sounding like your voice is underwater. Fix upstream first and the downstream tools barely have to work. That's the entire philosophy: every dB you remove at the source is a dB software never has to fake.

The order also protects your wallet. The instinct when a recording sounds noisy is to shop — a new mic, a paid denoiser, eventually a booth. But a measurement costs nothing, and it routinely reveals that the “noisy mic” was a gain knob, or that the room passes spec at 9 p.m. and fails at noon. Diagnose before you spend.

The six steps, in order

1. Measure your floor at working gain

   Record 20 seconds of silence exactly as you'd record a take — same mic position, same gain, same chair, breathing quietly. Read the RMS of that silence. Free tools do this in minutes; the full walkthrough is in how to measure your noise floor. Write the number down: it's your baseline, and every later step gets judged against it. Below −60 dB RMS already? Skip to step 6 and verify properly. Most untreated rooms land between −45 and −58.

2. Silence what can be silenced

   The cheapest noise reduction ever invented is the off switch. Refrigerator on a timer for the session (set a phone alarm so you turn it back on), HVAC paused, fans off, phone in another room, computer as far from the mic as cables allow. Then schedule: every house has a quiet shift, and in lived-in homes the floor commonly swings 8–15 dB between the loud hours and the small ones — our 24-hour floor protocol maps your own swing in a day. One line on physical treatment, because it's real but it isn't our beat: soft mass between your voice and hard surfaces beats any plugin applied afterward. That's the principle, whole and entire; we stay on the measurement and software side.

3. Set gain so the noise stays down

   Gain staging is where home VO is won or lost, because gain amplifies your voice and your noise floor by exactly the same amount — the only way to improve the ratio is to get closer to the mic, not louder in the knob. Target peaks around −12 to −6 dBFS while speaking at performance volume, and never “fix it in post” by recording quiet and boosting later: boosting raises the floor with the voice. The full method, including why OS-level input sliders should be left alone, is in gain staging for voice-over.

4. Work the mic like the instrument it is

   Technique is free noise reduction. At 15 cm from the mic your voice might arrive 20 dB above the room; at 60 cm the room catches up fast — inverse square law working against you. Get close (a hand-span is a good start), stay slightly off-axis to tame plosives, and keep your delivery energy consistent so your gain can stay put. A cardioid pattern pointed away from the noisiest direction — computer, window, door — buys real dB for free. Every one of these moves lowers how hard steps 5 and 6 have to work.

5. Close the gap with a light software chain

   Now, and only now, software. The standing order for recorded reads: a high-pass filter around 70–80 Hz to drop rumble your voice doesn't occupy, then gentle broadband noise reduction — think 6–12 dB of attenuation, not the slider's maximum — and nothing else unless a specific problem demands it. No gate slammed to digital silence: ACX QA listens for natural, continuous room tone, and gated-dead pauses sound processed. The architecture, presets by use case, and the live-versus-offline split are in the home VO software chain.

6. Verify against the spec, not your ears

   Ears adapt; meters don't. Re-run the measurement from step 1 on a finished file and check all three numbers — RMS, peak, floor — against the ACX checklist. If the floor still fails, don't reach for stronger software yet: identify what kind of noise is left, because hiss, hum, and rumble have different culprits and different cheap fixes. Then loop back to the step that owns the culprit. Two passes through this loop beat any single heroic plugin setting.

The two halves of the job — and the split that keeps you honest

Home VO noise is really two problems wearing one name. The recorded half — auditions, audiobook chapters, client deliverables — is solved offline, with the measured chain above, where you can undo anything. The live half — directed sessions, agent calls, coaching, callbacks — happens in real time, where a dog bark or a passing siren can't be re-taken and your professionalism is judged on the spot.

The live half is where real-time suppression like Krisp belongs: it sits between your mic and the call app and removes the household around your voice before the other side hears it. Keep it off your recorded masters — real-time suppression optimizes for intelligibility on calls, not for the natural room tone audiobook QA wants — and let it guard the conversations instead. The full where-it-fits breakdown is in Krisp for voice actors.

The recorded half

Auditions, audiobook chapters, client deliverables

Solved offline, with the measured chain — where you can undo anything

The live half

Directed sessions, agent calls, coaching, callbacks

Real-time suppression, sitting between your mic and the call app

What software cannot fix

Honesty section, because the ceiling matters. Software cannot remove your voice's own reflections — the boxy, roomy character of a bare-walled space is baked into the same frequencies as your voice, and “de-reverb” processing strong enough to remove it leaves audible artifacts on long-form reads. It cannot fix intermittent, voice-overlapping noise (a TV through the wall, conversation, a dog mid-sentence) without taking speech quality with it. And it cannot rescue a file recorded so quietly that boosting it lifts the floor above spec. When you hit those walls, the fix lives upstream: scheduling, positioning, gain — steps 2 through 4, run again. Knowing where the ceiling is keeps you from paying for promises above it.

The free path, start to finish

Worth saying plainly: every step of this method can be run for $0. Audacity records and measures; the community-built ACX Check plugin reads peak, RMS, and floor against spec; Audacity's own filters handle the high-pass and gentle reduction. None of those pays us a cent, and for a narrator watching the budget they're our first recommendation, not a consolation prize. Paid tools buy convenience and better edge cases — they don't buy your pass. The measurement guide runs the whole free path: measure your noise floor.

1. Audacity records and measures
2. ACX Check reads peak, RMS and floor
3. Audacity’s filters: high-pass, gentle reduction
4. Every step: $0

None of these tools pays BoothSense anything.

Questions narrators actually ask

Can I pass ACX in an apartment with street noise?

Often, yes — with scheduling and positioning doing the heavy lifting. Street rumble is low-frequency and steady, which makes it the easiest noise class to treat: a high-pass filter plus modest broadband reduction handles what closed windows and a smart recording time leave behind. Measure at three different hours before concluding your apartment can't do it.

Do I need a $500 denoiser plugin?

No. If your floor is within 10–15 dB of spec after steps 1–4, free tools close the gap cleanly. Premium tools earn their price on damaged or irreplaceable recordings — rescue work — not on a well-staged home setup recorded tomorrow.

Why does my audio pass the meter but still sound amateur?

Because the floor number doesn't hear character: clicks, mouth noise, plosives, and room reflections all live above the floor. ACX's spec explicitly requires files “free of extra sounds” on top of the numbers — we wrote up exactly this failure mode.

Should I leave noise suppression on all the time?

On calls, sure. On masters, no — record raw, keep every take unprocessed, and apply the chain in post where you can undo it. The one exception is a client who explicitly asks for processed delivery; get that in writing.

How long does this actually take?

First pass: an afternoon — ten minutes to measure, an hour of source-hunting and settings, an hour to build and test the chain. After that it's maintenance: re-measure when anything changes (new interface, new room, new season of HVAC) and before any big project.

Where to go next

If you only read one more page, make it the measurement guide — nothing else on this site works without your number. Chasing a specific gremlin, start at noise floor too high. Audiobook-bound, the full ACX checklist is the gate you'll be judged at, and the −60 dB deep dive explains the judge.