Audiobook Noise Floor: What It Is and How to Fix It

Noise floor is the third of the three ACX audio requirements (RMS, peak, noise floor) that cause automated rejection. ACX requires the level of background sound during silence to measure below -60 dBFS.[1] It includes everything your microphone picks up when you are not speaking: HVAC hum, computer fans, electrical interference, room ambience. The noise floor paradox, and why louder audio fails ACX: to meet ACX loudness requirements, you raise the gain. Raising the gain raises the noise floor by the same amount. If your raw recording sits at -63 dBFS and you add 6 dB of gain, your noise floor lands at -57 dBFS. That is a fail.

The Gain Budget: How Much Noise Can You Afford?

ACX's noise floor limit is -60 dBFS, measured in silent sections of the chapter file.[1] The 60 dB figure is not arbitrary: it is the level below which continuous background sound stops being audible to a listener over spoken narration at normal playback volume. At -50 dBFS, hum or hiss is clearly present between sentences. At -60 dBFS, most listeners cannot hear it even in quiet passages.

Every decibel of gain you apply during mastering moves the noise floor up by the same amount. The gain budget matters. If you need 8 dB of gain to reach the -23 to -18 dBFS RMS loudness window, your raw noise floor must start below -68 dBFS. The maths is inescapable. This is why narrators with clean, quiet recordings still fail after mastering. The raw file passed. The mastered file did not. The mastering raised everything, including the noise.

How to Measure Your Noise Floor

Measure after mastering, on the final MP3 you will submit. Select 2 to 3 seconds of silence (a natural pause between sentences, not the very start or end of the file). In Audacity, go to Analyse > Contrast and read the RMS of the background selection. In any other DAW, use the amplitude statistics or level meter on the selected silence. Measure at least three different silent passages and use the worst (highest) value. If that number is above -60 dBFS, your file will fail.[1]

Measuring noise floor in Audacity step by step

1. Open the mastered file in Audacity.
2. Find a silent section between sentences, at least 2 seconds long. Avoid the very beginning or end of the file, where silence padding may have been added artificially.
3. Click and drag to select the silent section.
4. Go to Analyse > Contrast.
5. Click Measure selection. The "Volume" reading for the selection is your noise floor measurement.
6. Repeat on at least two more silent sections in different parts of the file. Use the worst (highest) value.
7. If the reading is above -60 dBFS, the file will fail ACX's automated check.

For a quick sanity check before mastering: measure your raw recording's noise floor. Estimate how much gain you will need to reach -20 dBFS RMS. Add those two numbers. If the result is above -60, you know the mastered file will fail unless you manage the noise first.

Common Noise Sources in Home Studios

Most narrators record at home. These are the noise sources that push noise floors above -60 dBFS after mastering:

HVAC and air conditioning. The most common offender. Central heating, air conditioning, and fans produce continuous broadband noise between -55 and -45 dBFS depending on the room. Turn them off during recording. If you cannot (climate constraints), record during the quieter part of the cycle.

Computer fans. Desktop computers are loud. Even laptops produce audible fan noise when running audio software. Position your microphone as far from the computer as your cable allows, or move the computer outside the recording space. Some narrators record on a laptop with fans spinning up during the session. If possible, pre-cool the room, then record in short bursts before the laptop heats up.

Electrical hum. A 50 Hz or 60 Hz hum (depending on your mains frequency) from poorly shielded cables, ground loops, or nearby power supplies. Use balanced cables (XLR) between your microphone and interface. Keep audio cables away from power cables. A hum filter can help, but prevention is better.

Room reflections. Not technically noise floor, but reflections make noise reduction less effective because the reflected signal has a different frequency profile from the direct noise. Soft furnishings, blankets on hard walls, and commercial acoustic panels all help. The goal is not a dead room, just one where reflections do not colour the sound noticeably.

External noise. Traffic, neighbours, construction, weather. These are intermittent and harder to manage. Record during the quietest hours. Close windows. If your building has thin walls, a closet full of clothes is genuinely one of the best recording environments available.

What noise floor to expect from different environments

• Professional treated booth: -70 to -80 dBFS. Passes ACX comfortably.
• Closet with soft furnishings: -62 to -70 dBFS. Passes with modest gating.
• Quiet home office, HVAC off: -55 to -65 dBFS. Depends on how much gain is needed.
• Home office, HVAC running: -45 to -55 dBFS. Unlikely to pass without re-recording.
• Untreated room, open window: -35 to -50 dBFS. Will not pass.

Why Noise Reduction Often Makes It Worse

The instinct is to reach for noise reduction. Audacity audiobook mastering has one built in.[3] It works by capturing a noise profile from a silent section, then subtracting those frequency bands from the entire recording.

The problem: it subtracts from the entire signal, not just the silent parts. Your voice passes through the same filter. At moderate settings, this can sound acceptable. At the settings needed to pull a -57 dBFS floor down to -60 dBFS, it creates metallic, watery vocal artefacts that listeners notice immediately.

There is also the opposite problem. Aggressive noise reduction can push the noise floor too low. Passages that should contain natural room tone become unnaturally dead. Reviewers flag this as over-processing, and it can trigger a quality rejection even though the specs technically pass.

The approach is wrong, not the settings. Broadband noise reduction treats every sample the same regardless of whether it contains speech. That is the root of both problems.

Gating vs Noise Reduction

These are fundamentally different tools solving different problems.

Noise reduction analyses frequency content and subtracts noise bands from the entire recording. It alters your voice because your voice occupies some of the same frequencies as the noise.[3]

Noise gating reduces volume only during silent passages. When you are speaking, the gate is open and the signal passes through untouched. When you stop speaking, the gate closes and attenuates the silence.[2] Your voice is never processed.

A standard noise gate uses a fixed threshold: below a set level, the signal gets reduced. This works but creates problems when the recording has dynamic range. Quiet passages may get gated when they should not. Loud breaths may pass through when they should not.

Adaptive gating takes this further. Instead of a fixed threshold, it classifies each audio block based on its characteristics and adjusts the gating per passage. This distinction between speech, breath, and background noise is what separates a tool that works on chapter one from one that works on every chapter.

The Limits of Gating

Gating works on silent passages. It cannot separate noise from voice within the same audio block. If your recording environment is noisy enough that the background sound is audible under speech, improving the environment will give you better results than any post-processing.

A quieter recording space makes the biggest difference: close the window, turn off the HVAC, move away from the computer, treat the room with absorption.

For recordings that are clean but failing after mastering, the issue is almost always the gain budget. That is fixable. See fix ACX rejection for the step-by-step decision tree, or revisit the ACX audio requirements (RMS, peak, noise floor) for how noise floor sits alongside loudness and peak.

Common Questions

Your first file is free at ChapterPass.