How do you record an audiobook at home that passes ACX?
ACX's own numbers (RMS -23 to -18 dB, peaks below -3 dB, noise floor below -60 dB), how to keep a home room under them, what Audacity 4 changed for ACX Check, a mastering order with real settings, and a per-chapter checklist.
A 70,000-word novel is a big audiobook. ACX, the Audible platform where authors and narrators publish, says most performers narrate about 9,300 words per hour, so that manuscript comes out at roughly seven and a half finished hours. ACX also gives a budget for the work behind each finished hour: about 2 hours of recording, 3 hours of editing and 1.2 hours of listening back, 6.2 hours in total. Multiply it out and the book costs you around 47 hours at the desk.
Now picture losing those hours to one number you never looked at. You record twelve chapters in the spare bedroom, edit them, upload them, and the review comes back on noise floor: the hum of the fridge through the wall sits above ACX's limit in every pause. That is not an editing problem. It was in the room on day one, and every chapter inherited it.
This guide puts the steps in an order that catches that problem on day one, not after chapter twelve. It starts with what ACX actually measures, quoted from its own pages, then the room, the mic, the recording level, ACX Check in Audacity, and the editing and mastering chain, and it ends with a checklist table you can run on every chapter file. If you are still choosing hardware, the companion piece which microphone is best for voice over at home compares mics for exactly this kind of spoken-word work; come back here once you have one.
What ACX actually checks
ACX publishes its rules in one help article, the ACX audio submission requirements. The technical part fits on an index card:
- RMS between -23 dB and -18 dB. RMS is the average level of the file. ACX's own words: "Each file needs to fall between the specific volume range of -23dB and -18dB RMS."
- Peaks no higher than -3 dB. The loudest instant in the file has to stay 3 dB below digital full scale. ACX says the headroom also matters for encoding.
- Noise floor no higher than -60 dB RMS. This is the level of the file when you are not speaking.
- 192 kbps or higher, constant bit rate, 44.1 kHz MP3. ACX adds that 256 or 320 kbps is allowed but the difference listeners hear "will be negligible".
- All mono or all stereo. You can pick either, but every file in the book has to match. "Mixed formats will cause errors."
- Room tone at the head and tail. ACX recommends 1 to 5 seconds of room tone at the start and end of each file, and spacing "must not exceed 5 seconds".
Then there are the structural rules, which trip people up more often than the numbers:
- One chapter or section per file. Opening and closing credits are separate files.
- No file longer than 120 minutes. Split a long chapter and add a secondary header.
- Each file carries its section header ("Prologue", "Chapter 1"). Missing headers mean revisions.
- Opening credits state the title, author and narrator, matching the metadata and cover. Closing credits have to signal the end; ACX suggests "You have been listening to..." followed by the credits and "The End".
- A retail sample of 5 minutes or less, ideally the first five minutes, with nothing explicit in it.
- A human narrator unless Audible has authorized otherwise.
Two softer rules matter as much as the numbers, and no meter checks them. ACX wants the audio "consistent in sound and formatting", which it spells out as levels, tone, noise level, spacing and pronunciation. And it wants files free of "plosives, microphone pops, mouse clicks, excessive mouth noise, and outtakes".
There are also two layers of review, and they catch different things. ACX's own audio analysis tool and ACX Audio Lab check RMS, peaks, bit rate, bit rate method, sample rate, mixed channels, duplicate files and spacing. ACX states plainly that Audio Lab "does not detect issues such as noise floor or editing errors". ACX says those may be found later by its quality assurance team, after you submit. So a green result in Audio Lab tells you the file is encoded right. It tells you nothing about the fridge.
The room decides the noise floor
Of the three level numbers, RMS and peaks get fixed in mastering with two effects. The noise floor gets decided before you press record, because noise that sits under every word is hard to remove without changing the voice along with it.
A home noise floor comes from three places:
- Machines. Computer fans, air conditioning, the refrigerator, a ticking clock. An ACX producer interviewed on the ACX blog warns against recording with a laptop near the mic, or anything else "with a cooling fan that turns on and off as the temperature fluctuates". The on-and-off part is the trap: a fan that is quiet during your level check and spins up in minute twenty gives you a chapter whose noise changes halfway through. If you narrate on a MacBook Air, Apple describes it as having "a silent, fanless design", which removes one source outright. A MacBook Pro or a desktop Mac has fans, so move it as far from the mic as the cable allows, and never block its vents to quiet it; the Audacity audiobook guide says the same.
- The building and the street. Traffic, dogs, neighbors, planes. The Audacity guide is blunt about these: there are no tools to remove noises that constantly change, gating does not help, and the fix is to "read the work again in a quieter room". Pick your recording hours around them.
- The gear itself. The same ACX article quotes a producer who soundproofed his room and then found "it was the cheap mic that I was using that was generating most of my noise floor." Hiss from a mic or preamp is steady, so it responds to gentle noise reduction, but it is better not recorded at all.
Then there is echo. Bare walls give a voice a hard, roomy tail. No ACX meter measures it, but a listener on headphones hears it. You do not need a booth. Another contributor to that ACX piece, stuck in a noisy five-story apartment building with window air conditioners, describes hanging packing blankets and winter coats in a closet and recording in there. A closet full of clothes works for a simple reason: it is soft on every side and small, so reflections die fast. If there is no closet, hang heavy blankets behind and beside you, close the curtains, and put a rug under the desk. For the difference between keeping sound out and stopping it bouncing around, see whether you can soundproof a room for recording; for a home narrator, absorption near the mic usually matters more than isolation.
Measure before you commit a weekend. The free mic test on this site records 2.5 seconds of silence and 10 seconds of reading in your browser, then reports loudness, noise relative to your voice, the echo tail and the voice band, and plays the take back. The noise meter shows how loud the room is over time, which is how you catch the fan that cycles every few minutes. Run both in the spot where you plan to record, at the hour you plan to record.
Dynamic or condenser for an untreated room
The usual picture of audiobook recording equipment is a large condenser mic in a padded booth. That picture is right for the booth. In an ordinary bedroom it often backfires.
A condenser capsule is sensitive. It picks up detail in the voice, and it picks up the room just as faithfully: the echo off the window, the laptop, the street. A dynamic mic needs more gain from the preamp and you have to work closer to it, but at that distance your voice is much louder than the room, so less of the room ends up in the file. The trade is real: more gain means more preamp hiss, so a dynamic mic on a weak interface can raise the noise floor from the other direction.
The position this guide takes: in an untreated or half-treated room, start with a dynamic mic, close up; move to a condenser once you have a treated closet or corner. The best microphone for audiobook narration is the one that keeps the room out of your pauses, not the one with the widest frequency chart. The longer argument, with the physics, is in dynamic or condenser for a normal room.
Distance matters as much as the mic type. The Audacity guide suggests being "about a shaka away", the width of a hand with the thumb and little finger spread, or as close as a fist for a louder, more intimate sound, where you will probably need a pop filter. Whatever you choose, keep it identical for the whole book. Mark the mic stand position and your chair position with tape. A narrator who drifts from a fist to a hand-span between Tuesday and Thursday changes the tone of the book, and ACX lists tone under consistency.
One more trap for USB mics on a Mac: the Audacity guide describes a "mosquito whine" that some computer and USB mic pairs produce, and the ACX Check plugin page warns that a 1000 Hz whine from USB audio "may result in an ACX rejection even though it is below the -60 dBFS noise floor requirement". Listen to a pause on good headphones before you record a whole chapter.
Record at a safe level, and keep the raw take
Recording too hot is the mistake you cannot undo. A clipped word stays clipped. The Audacity guide's advice is to record to "a reasonable volume (occasional peaks at -6dB) in a quiet room with no echoes". That leaves room above your loudest line, and mastering lifts the average later. If you keep hitting 0 dB, turn the input gain down, not the output; gain staging explained walks through where each knob sits.
A few settings that save rework:
- Sample rate 44.1 kHz. ACX requires 44.1 kHz files, and ACX Check warns when a selection is at anything else. Recording at 44.1 kHz from the start avoids a conversion step.
- One mono track. One mic is one channel. If you later export stereo, export every file stereo.
- Uncompressed while you work. Record and edit in WAV or the editor's own project format. MP3 is only the final delivery format.
- Keep a raw backup. Before you touch a chapter, the Audacity guide recommends exporting a WAV of the raw reading and keeping it "ideally not on the machine", so an editing accident never means reading the chapter again.
- Record room tone. Ten seconds of the room with you sitting still, at the start of each session, gives you clean material to fill gaps and to build the 1 to 5 seconds at the head and tail of each file.
ACX Check in Audacity, in late 2026
ACX Check is a free analysis plugin for Audacity written by Steve Daulton, based on an earlier plugin by Will McCown. It is a Nyquist plugin, a single ACX-Check.ny file you download from the Audacity plugins site and install. Select a chapter, run it, and it shows three numbers against ACX's limits:
- Peak level, the highest peak in the selection.
- RMS level of the selection.
- Noise floor, measured as the RMS of the quietest 500 milliseconds in the selection.
It also warns when the sample rate is not 44,100 Hz, when the selection runs longer than 120 minutes, and when the noise floor reads as minus infinity, meaning dead digital silence, which "sounds unnatural". ACX wants low, constant room tone, not silence, so do not paste digital zeros into your pauses.
The plugin's own page is frank about its limits. Even a clean pass "is NO guarantee of ACX acceptance", and the noise floor is taken from the quietest half second of the selection. That second point has a practical consequence: one unusually quiet moment can make the number look better than your real pauses, so check a stretch of normal pauses, not only the room tone at the head.
Where it lives depends on your Audacity version, and this changed recently:
- Audacity 3.x. The official audiobook mastering guide is written for it. After installing the plugin you run it from Analyze > ACX Check. The last 3.x release on the older-versions page is 3.7.9.
- Audacity 4. Version 4.0 shipped on 3 September 2026 and 4.0.1 on 30 September 2026. It rebuilds the interface on Qt, and its release notes say Nyquist plugins are supported, that "some controls have moved or changed", that the Macro Manager is not available in 4.0, and that 4.0 ships with some analyzers and effects missing. Neither the release notes nor the mastering guide, which is written for 3.x, gives a 4.x menu path for ACX Check or for the chain below. On the Mac download page, the Apple silicon (ARM64) DMG is marked "No plugin support", so if you want ACX Check in 4, install the universal DMG.
If you want the documented path with the least guessing, use Audacity 3.7.9 for mastering: the guide, the menu paths and the optional mastering macro all match it. For the rest of the setup, setting up Audacity on a Mac covers input devices and recording.
Edit first, then master in a fixed order
Audiobook mastering is short. Editing is long. ACX's own budget gives editing more time than recording, and that time goes on the things no meter catches: misreads, retakes, lip smacks, page turns, a chair creak.
Edit each chapter first. Cut the outtakes, tidy the breaths that are louder than the words around them, remove clicks and pops, and set the head and tail room tone to the same length on every file. Then master, in this order.
1. Noise reduction, only if the noise check fails, and lightly. The Audacity guide suggests starting gentle with its Noise Reduction effect (settings 6, 6, 6, then 9, 6, 6) and warns that if you need as much as 12, 6, 6 "your voice may get wine-glassy or honky and the show may not pass ACX inspection". Our guide to removing background noise from a voice recording covers what to do when the noise is not steady.
2. EQ. ACX's engineering column recommends applying EQ before the level steps, because unwanted low and high frequencies affect the limiter that comes next, and cutting them also tames hum and hiss. In the Audacity guide, a low roll-off for speech is the first step.
3. Level: normalize, compress or limit. ACX describes this step as bringing files to the required levels "by using normalization, compression and/or limiting, and, if necessary, a final volume adjustment". Its worked example: a file peaking at -4 dB with an RMS of -27 dB gets a limiter that pulls the peaks down by 3 dB, which then leaves room to raise the whole file by 4 dB, landing at -23 dB RMS with peaks at -3 dB.
4. Check after every pass. Run ACX Check after each effect, not only at the end, so you know which step moved which number.
For Audacity 3.x, the official guide gives a concrete chain with real settings. Treat it as a starting point, not a law:
Effect > EQ and Filters > Filter Curve EQ
Factory Presets > Low roll-off for speech
Effect > Volume and Compression > Loudness Normalization
Normalize RMS to -20 dB
Effect > Volume and Compression > Limiter
Soft Limit, 0.00, 0.00, -3.50 dB, 10.00, No
Analyze > ACX Check
target: peak below -3 dB, RMS -23 to -18 dB, noise below -60 dB
The guide says the peak and RMS readings after this should be "nearly perfect", and if the noise is quieter than about -65 dB and the chapter sounds reasonable, you may be done. Its settings are sticky, so after the first chapter, mastering each new one comes down to select, apply, apply, apply, check.
Then listen. ACX's own mastering checklist ends with exactly this step, and gives the test: if the result sits at one unchanging volume, it is over-processed; if it jumps and drops, it is under-processed. The Audacity guide adds the warning that matters most for home narrators: "It's also possible to pass ACX-Check and still fail acceptance." Heavy processing fails ACX's human quality control, and the guide names that failure "Overprocessing". Gentle and consistent beats perfectly flat.
Export last: MP3, constant bit rate, 192 kbps or more, 44.1 kHz, the same channel format for every file. ACX points to the free encoder fre:ac and its own encoding tutorial if your editor's MP3 export gives you trouble.
The per-chapter checklist
Run this on every file before upload. The middle column is what ACX asks for; the right one is how you check it at home.
| Check | ACX target | How to check |
|---|---|---|
| RMS | -23 dB to -18 dB | ACX Check, or ACX Audio Lab after upload |
| Peak | no higher than -3 dB | ACX Check, or Audio Lab |
| Noise floor | no higher than -60 dB RMS | ACX Check on a stretch of pauses (Audio Lab does not measure it) |
| Format | MP3, CBR, 192 kbps or higher, 44.1 kHz | export settings, Audio Lab |
| Channels | all mono or all stereo | export settings, Audio Lab |
| Head and tail | 1 to 5 seconds of room tone, never more than 5 | look at the waveform; Audio Lab checks spacing |
| Length | 120 minutes or less per file | ACX Check warns |
| Content | one chapter or section per file, with its header | your chapter list |
| Credits | opening and closing credits as separate files | your chapter list |
| Sample | 5 minutes or less, nothing explicit | your sample file |
| Clean audio | no plosives, pops, clicks, mouth noise, outtakes | listen on headphones |
| Consistency | same level, tone, noise, spacing, pronunciation | play the first and last chapter back to back |
The last row deserves its own habit. Before you upload, play thirty seconds of chapter 1 and thirty seconds of the last chapter one after the other. Home recordings drift over weeks: a different time of day, a cold, a mic that crept closer. You hear it in a minute, and a listener hears it too.
Where TunedMic fits, and where it does not
TunedMic is a paid Mac app that adds a virtual microphone to macOS. Your real mic stays as it is; TunedMic processes its sound on the Mac in real time, and any app that lets you choose an input, Audacity included, can record from it. Setup is a Sound Check: you read a short passage aloud for about a minute while it measures your voice, your mic and your room, and the profiles are built from that. It runs locally and adds less than a tenth of a second of delay. The 38-second video on the home page shows the idea on a call: a hair dryer, a humming laptop and a crackling headset, then one click leaves only the voice.
For an audiobook that cuts both ways, and it is fair to say both.
The case for it. It cleans the room sound before it reaches the recording, so the file you edit already has less fan and less echo in it. For a narrator who cannot build a closet, that can be the difference between chasing the noise floor in every chapter and not.
The case against it. ACX wants consistent, natural audio, and its human reviewers reject over-processing. Anything processed while you record is baked into the file; you cannot undo it later the way you can undo a noise reduction pass. That is the argument for recording raw and doing all processing afterwards, where every step can be redone. Both approaches are legitimate. If you try TunedMic for narration, record the same page both ways, run ACX Check on both, listen on headphones, and decide with your ears. If you use it, use one profile with the same settings for the entire book, because switching profiles changes the sound everywhere at once, and a change in tone between chapters is exactly what ACX flags. And to be plain: TunedMic does not guarantee that a file passes ACX. Mastering and the checklist above still apply.
The more clearly useful part for narrators is the Speech Report, included with the license. Turn the report on, use Start Practice, read a few pages of the next chapter as a rehearsal, and you get the same report TunedMic gives after a call: your pace, your pauses, the longest stretch you talked without one, how often you said "um", and how lively your intonation was, each compared with your own last 20 sessions. Filler-word counting is optional and needs a one-time download of a speech model of about 0.5 GB. Only the numbers are kept, never audio or a transcript, and nothing leaves the Mac. A rehearsal that shows your pace creeping up or your "um" count climbing tells you to rest before you record, which is cheaper than an hour of retakes. The Speech Report needs macOS 14.4 or later.
TunedMic needs a Mac with Apple silicon (M1 or newer) and macOS 14 or later. There is no Windows or Intel version. If that is not your computer, everything else in this guide still works: Audacity and ACX Check run on Windows and Linux too, the closet and the blankets work anywhere, and the browser mic test and noise meter run on any OS. If it is your computer, the free 7-day trial turns on every feature and needs no card, and pricing lists the one-time license.
What to do tonight
Do not record chapter 1 yet. Record a test instead, in the exact spot and at the exact hour you plan to narrate:
- Ten seconds of room tone, sitting still.
- One page of the book, read the way you will read the whole thing.
- Run ACX Check on the whole take, then on a stretch of normal pauses.
If the noise floor reads higher than -60 dB, fix the room before anything else: move the computer, kill the fan, try the closet, try a later hour, and test again. Only when that number sits comfortably below the limit does it make sense to record a full chapter, master it with the chain above, and upload it to ACX Audio Lab. One chapter that passes becomes the template for the rest of the 47 hours.