ACX requires that each audiobook file has "a noise floor no higher than -60dB RMS to avoid background noise distractions." Of the three ACX level rules, it is the one mastering can do least about: RMS and peaks are set with gain and limiting, but the noise floor is mostly decided when you record. This guide explains how the floor is measured, why levelling raises it, and how to get under -60 dB without the over-processed sound that causes problems of its own. For the full rule set, see ACX audio requirements explained.
What the noise floor is
The noise floor is the level of everything in the recording that is not the narration: room noise, electrical hiss, air conditioning, distant traffic. You hear it in the pauses between sentences, and it is still there underneath the words. With narration inside ACX's -23 to -18 dB RMS window, a floor at -60 dB sits roughly 37 to 42 dB below the voice.
How it is measured
ACX does not publish a measurement method for the noise floor, and its Audio Lab analyzer does not check it; the floor is judged by people during ACX's quality review. The closest objective reference is the ACX Check plugin for Audacity, published by the Audacity team.
The plugin's documentation describes the noise floor as the RMS level of the quietest 500 milliseconds in the selection. Its code is a little more precise: it computes the RMS of a 400 ms window, moves the window along in 100 ms steps across the whole selection, and reports the lowest value. A reading above -60 dB fails. A reading below -90 dB triggers a warning, because dead silence sounds unnatural.
Two practical consequences follow:
- One clean pause is enough for the plugin. It reports the quietest window, so the room tone at the head or tail of a file usually sets the reading. ACX's reviewers listen to the whole file, though, so a clean head with a noisy middle can still be flagged.
- Measure the final file. The floor of the raw recording is not the floor you deliver.
You can measure a chapter in your browser with the free ACX check, which uses the same 400 ms window and reports stretches of pure digital silence separately.
Why levelling raises the floor
Gain applies equally to voice and noise. Suppose raw narration measures -26 dB RMS with a floor of -64 dB, so 38 dB separates them. Raise it to -20 dB RMS, a common target in the middle of the ACX window, and the floor rises by the same 6 dB to -58 dB. That fails, even though nothing about the room changed.
Compression makes it worse. It narrows the difference between loud and quiet parts and makes up the level with gain, so quiet material, including the noise in the pauses, ends up closer to the speech. What matters is the distance between narration and noise after processing. If the narration will end up around -20 dB RMS, you need at least 40 dB between the two, plus a few decibels for safety. Audacity's audiobook mastering guide makes the same point: after normalizing to -20 dB RMS and limiting, a floor quieter than about -65 dB may need no further work.
Fix it at the source
Every decibel you do not record is a decibel you do not have to remove. The main fixes, covered in detail in how to reduce background noise in voice recordings, are:
- Record in a small room with soft furnishings, away from windows and exterior walls.
- Work close to the microphone: halving the distance raises your voice by roughly 6 dB relative to the room.
- Switch off heating and air conditioning, keep fridges out of the room, and move noisy computers away during takes.
- Put phones in airplane mode and track down any electrical hum.
- Measure a short test recording at the start of each session, before the long takes.
Room tone, not silence
ACX recommends 1 to 5 seconds of room tone at the beginning and end of each file, and room tone spacing must not exceed 5 seconds. Room tone is the sound of your recording space with nobody speaking. It should match the background under the narration, so record it in the same session, with the same microphone position and gain, and use it to fill edits.
Do not replace room tone with digital silence. When pauses drop to absolute zero, the background under the voice becomes more noticeable each time the narration returns, and the ACX Check plugin warns when the floor falls below -90 dB. The aim is a quiet, natural, consistent background, not no background at all.
Processing that helps
Rumble filtering
A high-pass filter around 80 Hz removes low-frequency rumble from traffic and ventilation. An unweighted RMS reading counts rumble in full even when it is hard to hear, so removing it can lower the measured floor without touching the voice.
Spectral noise reduction
Spectral noise reduction lowers steady noise such as hiss and air-conditioning rush, in the pauses and, to a degree, under the voice. Apply it before compression and levelling, and use the least amount that does the job. A moderate setting can move a borderline floor by a few decibels; heavy settings create artifacts that are easy to hear on headphones.
Gentle downward expansion
A downward expander turns the level down by a limited amount whenever the signal falls below a threshold, which is what happens in the pauses between phrases. Unlike a hard gate, it lowers the noise instead of removing it, so the pauses keep a natural room sound. Keep the range modest, and use a release slow enough that word endings and breaths are not chopped off.
Signs of over-processing
- A watery, metallic or "underwater" quality on the voice.
- Word endings, soft consonants and breaths that sound cut short.
- Background noise that audibly switches on with the voice and off in the pauses.
- Pauses that sound dead next to the room tone at the head of the file.
- Chapters with audibly different backgrounds, which breaks ACX's consistency requirement.
A large gap is a recording problem. Processing can close a few decibels cleanly. If the floor is far above -60 dB, a quieter room or a closer microphone position will cost less in sound quality than heavy noise reduction.
Putting it together
- Record a short test with room tone and measure its floor before the session.
- Edit, filling gaps with room tone rather than silence.
- Apply a rumble filter and light noise reduction.
- Level and limit into ACX's RMS and peak limits.
- Measure the floor on the final MP3, then listen to the whole chapter on headphones.
For ACX projects, LufsReady's ACX audiobook preset measures the noise floor the way the ACX Check plugin does, applies medium spectral noise reduction (12 dB) by default and, only when the floor would otherwise miss the limit, adds a silence-cleanup expander (which you can switch off) that lowers room noise between phrases by a limited amount instead of cutting to digital silence. The report shows the floor before and after processing and notes any digital silence it finds. It cannot make a noisy room quiet, so the recording fixes above still come first.
Sources
- ACX: Audio Submission Requirements
- ACX: Audio Lab (audio analysis tool) FAQ
- Audacity: Analysis plugins (ACX Check)
- Audacity: Audiobook mastering guide
Check your file in seconds
Measure loudness, true peak and noise floor for free in your browser, or master a file to the exact target with a report.