AudioBook Mastering version 4

Audiobook Mastering is now a formal part of the Audacity Manual.

This forum posting is an early work in progress and no longer updated or corrected.

AudioBook Technical Mastering

(revision 1, minor corrections 2017-12-08)
(revision 2, New and Improved RMS Normalize 2018-07-25)
(revision 3, Bold for clarity 2018-08-25)

This is a suite of Audacity tools and process to help you produce an audiobook reading that reaches ACX Technical Compliance. See NOTES AND COMMENTS at the end.

We designed custom AudioBook Mastering tools. You have to install them.

Get and install ACX-Check

Get and install RMS-Normalize

Use Audacity version 2.1.3 or later. Not optional.

Start this process with a raw, uncorrected reading. No processing, adjustments or fixes. Don’t help.

Export a WAV (Microsoft) 16-bit sound file of your raw reading and save it in a safe place. You should never be stuck reading a chapter again because of an accident. Saving an Audacity Project is not recommended for this.

These instructions are in short-form: Location > Tool: Options > OK

Select the whole reading or chapter by clicking just right of the up arrow button (on the left).

Effect > Filter Curve > Manage > Factoy Presets > Low rolloff for speech > OK.

Effect > RMS Normalize: Target RMS Level -20dB > OK.

Effect > Limiter: Soft Limit, 0, 0, -3.5dB, 10, No > OK.

Analyze > ACX-Check.

The first two readings, Peak (no louder than -3dB) and RMS (between -18dB and -23dB) should be nearly perfect. If the noise is quieter than about -65dB (-60dB limit), and the show sounds reasonable, you may be done.

If you fail noise, then the process can be a great deal more interesting. See Noise Posting, below.



ACX has their own help pages, tutorials and videos.

This mastering suite addresses technical considerations. Nothing here will address theatrical errors such as harsh sibilance, gritty voice or wet mouth noises. There is no acting filter. You have to know how to read aloud.

The quality of the original recording is very important. It’s easy to record mistakes that can’t be fixed later. Record to a reasonable volume (occasional peaks at -6dB) in a quiet room with no echoes.

When you get done with a chapter, File > Export the work as WAV (Microsoft, 16-bit) and save the sound file with a unique name somewhere safe. That’s your emergency backup so you don’t have to read it again if you damage the work or lose your edit.

Only then start editing, processing, filtering and correcting. Keep testing the work with the add-on ACX Check analysis tool as you go. Select the work and Analyze > ACX Check.

ACX Technical limits:
— Peaks no louder than -3dB.
— RMS (performance loudness) between -18dB and -23dB.
— Noise no louder than -60dB.
Those are the top three values in the ACX Check panel

ACX Values in English

Human Quality Control at ACX (the theatrical test after you pass ACX-Check technical test) does not like heavy processing. You should be as gentle as you can with as few corrections as possible. Don’t even think of submitting readings that sound like a bad cellphone, speaking into a wineglass or reading in a bathroom.

The AudioBook metaphor is listening to someone telling you a story over cups of tea. Anything that distracts from that ideal should be avoided.


Noise Compliance – (revision 1, minor updates 20171208)

We assume you’ve already been through AudioBook Mastering and failed ACX Technical Compliance because of noise (louder than -60dB) or your performance has odd background sounds you don’t like. Noise is common in a home studio.


This is a gentle spring rain in the trees ffffffffff sound behind your voice. Microphone systems make noises like this naturally and it’s your job to make your voice loud enough so nobody notices the noise, but not so loud your voice distorts.

If the hiss isn’t too bad, try a gentle correction such as Noise Reduction of the Beast (6, 6, 6) or UNDO and try 9, 6, 6. You can hear the hiss getting quieter and recede into the background as you increase the first number: Nose Reduction. Try ACX-Check. If you need reduction as high as 12, 6, 6, your voice may get wine-glassy or honky and the show may not pass ACX human inspection.

The solution may be to change your announcing style or even the microphone. You should be about a shaka away from the microphone…

… or as close as a fist (louder and more intimate) but you may need a pop and blast filter.

Computer fan, air conditioning, refrigerators or other machines.

Turn off fans or machines while you’re presenting if you can. And yes, we understand the contradiction of needing to watch the Audacity screen, remove the noisy computer from your room and keep the computer one USB cable away from your USB microphone all at the same time. ACX did it by having an almost completely silent laptop in the studio to do the recording.

People have done it by extending the keyboard, mouse and monitor outside their studio, but you are warned against extending a USB cable, particularly with audio (or video) production.

Do not block computer vent holes.

If machine noises are constant and not seriously loud, they may respond to Effect > Noise Reduction.

Try a gentle correction such as Noise Reduction of the Beast (6, 6, 6) or UNDO and try 9, 6, 6. You can hear the hiss getting quieter and recede into the background as you increase the first number: Nose Reduction. Try ACX-Check. If you need reduction as high as 12, 6, 6, your voice may get wine-glassy or honky and the show may not pass ACX inspection.

Some computers and USB microphones hate each other and produce mosquito whine sound.

Whine doesn’t respond well to Effect > Noise Reduction.

Get and install Mosquito-Killer4.

Effect > Mosquito-Killer4: …Mosquitoes to kill?: 8 > OK.

We expect the tool to make common USB whine vanish, but there are some versions of whine that don’t perfectly respond. Listen carefully and if Mosquito_Killer4 fails or doesn’t work enough, UNDO and post a help message on the Audacity Forum with a sample of the work.

I don’t know of any easy, foolproof way to permanently fix a USB microphone once you have the whine sound. The most likely fix is change the computer. You can do it by not using a USB microphone. You can use a very high quality analog microphone plugged into a stable USB interface.

There are no tools to remove noises that constantly change. If jets overhead, traffic noises, dogs barking and the TV next door are included in your show, they are now your permanent performance partners. Read the work again in a quieter room. Gating is very difficult to use and generally doesn’t help.

Nobody said you can’t have more than one noise. The ACX AudioBook noise test can be rough to pass with a home recording system. If you just can’t get there with the above tools, or you can’t get your voice to sound right, post to the Audacity Forum with a sound sample and send a clean, raw clip. Mention the three-digit Audacity number.

Raw, please. We can’t remove effects from a clip and posting your correction errors doesn’t tell us anything.


When I run through the steps you’ve laid out I notice that nothing seems to happen (no progress bar, change in wave forms etc) when I select RMS Normalize from my effects - and my ACX Check values show my RMS is in the 40s but everything else is passing.

What silly thing have I overlooked?

dottedllama, are you sure you have selected the entire track you are working with first? Click in that area to the left where it says “Mono, 44100Hz, 32-bit float” and the track should be selected. The report you display shows about an 18 minute selection was measured, so it looks like you did. Other than that, I’m not sure why RMS wouldn’t change anything. That’s usually the one that produces the most obvious change in the waveforms in my track, as it makes everything louder.

What silly thing have I overlooked?

RMS Normalize requires Audacity 2.1.3 or later.

We went around the barn several times recently when, like you, the reader had the tools fail and nobody could figure out why.


Ah you see? I told you it was something ridiculous. Off to upgrade my version of Audacity…

Thanks Koz - as always you help keep me from leaving my head in my bum for too long!

Edited the original post, I had the files from book 3 up and they are pushing back on book two. So links and acx check screen shot reflect book two now. Thanks

I did my last audiobook with the sage advice and awesome help from here. So I was all like “yeah I got this” because it came out really well and sailed right through ACX checking and into the store for sale.

Now I did books 2 & 3 in the series and submitted them and ACX said there’s a noise floor issue (how dare they). I did everything the same, same environment as last time, same gear, etc. I wonder if I missed a setting or maybe wasn’t close enough to the mic? I’ll post links to the raw WAV and resulting MP3 after I did the recommended EQ, RMS Normalize, and Limiter effects, and the resulting ACX Check result. To be fair, I do hear a low hiss. I’ll try Noise Reduction of the Beast and the recommendations around NR but wanted to see if anyone could tell me what I did wrong this time around so I don’t repeat it.

But why are they pushing back if my file says it passes acx check? This is just one chapter but they said it’s all of them (all have similar numbers, all pass).

Thanks for all of the excellent help here!

Raw WAV:

Processed MP3:

But why are they pushing back if my file says it passes acx check?

Did they give you a reason? They almost always give you a narrative of why it failed. Just telling someone their dB noise value didn’t pass isn’t useful.

A note here, ACX-Check isn’t published by ACX. It’s a tool designed by flynwill and company to “simulate” the ACX Robot. The Robot gets your work before a human sees it with the idea if you fail that, they don’t have to go any further. I’m really interested in their fail post.

It’s most unusual to pass ACX Check and still fail submission.


The ACX Check plug-in is intended as a “guide”, not a “gold standard”. It is a relatively simple tool. The noise floor measurement makes lots of assumptions and so does not always give accurate measurements. I’ve run some tests on the ACX Plug-in, and found that the noise floor measurement can sometimes be a long way off. More important than any noise floor measurement is what it sounds like.

the noise floor measurement can sometimes be a long way off.

Which way and why?
I know the plugin needs a continuous half-second of room tone to measure noise. If the performer doesn’t leave a half-second, the tool will guess at it.

Also, that Low Rolloff thing is a big deal. Can’t leave that out.


It may indicate a much lower noise floor than the actual noise floor (it’s a bug).

The two provided sound samples will not play (for me) and they download severely damaged or empty.

It may indicate a much lower noise floor than the actual noise floor (it’s a bug).

Is there a method of cross-checking the work if you encounter an apparent conflict? ACX Check is a formal design of a much larger set of individual tools and processes you had to use to do this check manually in days gone by.

Does Analyze > Contrast give a rational answer, or that’s where the bug is hiding?

Is there an “unbug” we can borrow and slide into ACX Check v2?

I’m on the edge of my seat to read the ACX Rejection comment.


It was from a guy named Brendan Solorz:
“I have reviewed the resubmitted files for this title but found that they still contain issues that need to be resolved before this title can be moved forward. The submitted files have noise floors that are higher than required. For more info on noise floor, click here.”

Let me relink:
raw wav:
processed mp3:

They shockingly accepted book 3 which I thought might be worse. I did NR 9,6,6 on these from book two and they sound much better. I resubmitted so I’m on the edge of my seat now :slight_smile: Still not sure how the noise crept in since I thought everything was the same from last time. that’s what I’m really interested in.

I want to inspect and listen to the raw clip when I get home.

Noise is a little magic. That’s why I left it to last in the correction suite. Noise has pages of explanation and notes all on its own.

If your work is right on the edge—most home-style microphones work this way—just leaning back during the reading is enough to send you into violation. The grill of the microphone is where the rubber meets the road. The microphone internal noise is usually fixed and it’s your job to make your voice 1000 times louder (60dB).

But you can’t get too close because the microphone will start to pick up every tick, pop and slobber.

Noise Reduction isn’t a gift from the angels, either. High applications tend to boost ticks, pops and slobbers. It is strongly recommended that you do the same thing to each chapter even if some of them seem not to need the corrections. ACX puts a lot of stress on chapter to chapter matching.



I processed your submitted raw WAV and converted to MP3 submission standard. That room tone after the introduction came in with multiple different techniques at -73.6dB. Well within standards. No noise reduction.

The same patch on your MP3 posting arrives at -69dB, some 5-1/2 dB louder than mine. 6dB is double. I’m guessing from the text that’s with 9, 6, 6 noise reduction? That means you started with -60dB noise.

Where did all that noise come from? Unless I’m missing what you did.

I left all the false starts and tick marks in there. The new mastering suite doesn’t care about those unlike the old process.

I’m lost. What processing did you do between the Raw reading and the conversion to MP3?


There may be a mistake there. I’m juggling too many Audacity windows.

Still, do you apply any filters or equalizers (beside Low Rolloff)?


I didn’t do any processing to the raw file as far as effects. I probably chopped off the beginning and ending “setup noises” to get a shorter/cleaner track and I might have copied my one second of room tone to the beginning, section between 'chapter name and start of content, and end. That’s what I usually do before saving the .wav raw file.

The MP3 only has the EQ, RMS normalize, and Limiter effects (at the recommended values) done. No NR has been done on it.

Anyhoo, I applied the NR to every chapter and it does sound better. I submitted it like that, so let’s see. I’m pretty sure I was just leaning back too far this time. Lesson learned!

You don’t have to guess at it. You can make the Audacity sound meters larger so you can see them as you record (like the recording engineer would be doing).

That’s my standard meter setup. Note on the left, it’s not -60 or -72, but -96, the full volume range of 16-bit audio. Then I stretched the meter the whole way across the screen so the common recording voice range is visible on the right. -6dB on the meters works out to be 50% on the blue waves, and you can make the blue waves follow you as you record.

The elements of the top of Audacity are all adjustable by clicking, pulling and dragging. Click in the ribbed bar on the left and pull, or some windows will let you click both left and right.

I know someone is going to ask about watching the meters and text as the same time.

Yes. The only other way to do this automatically is AGC or Automatic Gain Control. That can produce weird effects and pumping background noise. It can get you flagged by ACX as Overprocessing.

If your meters are big enough, you an see the bouncing lights start to turn yellow without directly looking at them. That was not a design accident.


One happy accident of the mastering suite was the ability of the tools to gracefully adjust for minor variations in voice volume. If you have “golden ears” and are really paying attention, you might be able to find one phrase I voiced a bit louder than the others. But it will probably slide right on by.

It won’t compensate for leaning forward and bellowing into the microphone, but it does a dandy job with natural voice variations—without seeming to do anything.

By the way, In My Opinion, I think ACX made a mistake. Every third breathing human on earth has decided to read for audiobooks so their workload must be nuclear. I think they made a fluff and if you submitted the middle book again it would go straight through. You recorded Two and Three with basically the same characteristics and yet only Three made it through.

None of us can figure out what happened, so this is the most likely path solution. Occam’s Razor, et al.


Great info, Koz. The screen shot is really helpful. One thing is that I’ve been keeping the laptop just outside the closet door on a stand, so I can’t see the meters. But in thinking about it, my laptop (Asus Q552UB) is pretty quiet, I’m not even sure it has a fan, although the hard drive does make low ticking noises sometimes. Maybe next time I’ll do an A/B test with it in the closet with me so I can watch the meters.

I also realize that for books 2 & 3 I wasn’t recording with the headphones on, I bet that made a difference as far as my voice level.