My Audiobook Process for Removing A/C Fan Noise

[Zash]

Hello all,

I would like your thoughts on my process and results.

I am recording in my office. The main issue I have is that it is summer and there are booster fans running for the a/c system. I am not able to turn these off. I am uploading some clips below to get your thoughts.

Here is the Spectrogram of the raw WAV and Spectrum of 2 seconds of room noise:

The fan noise appears to be at about 713 Hz.

With the mic recording level at 100% and the Samson Q2U oblique positioned, I get amplification of about 9 dB.

Here is my process:

---

Before running macro:
0a. Manually edit out clicks and pops (I haven’t done much of this)
0b. Ctl-a, Analyze, ACX Check

0c. Effect, Noise Removal and Repair, Noise Reduction, Get Noise Profile <— HAVE TO DO THIS!!! *****

MACRO “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX”

0. Select All, Zoom to Selection
1. Noise Reduction, 18,6,6… or 12,6,6… or Beast 6,6,6
2. Effect, EQ & Filters, Filter Curve EQ, Presets & settings, Factory Presets, Low rolloff for speech, Apply
3. Effect, Volume and Compression, Loudness Normalization, Normalize RMS to -20.0 dB
4. Effect, EQ & Filters, Low-Pass Filter, Freq 6300 Hz, Roll-off 24 dB/octave
   …Note: I had to move Low-Pass Filter before Limiter to avoid Peak Level sometimes being too high
5. Effect, Volume and Compression, Limiter, Soft Limit, 0.00, 0.00, -3.50, 10.00, No, Apply
6. ACX Check

Then export: WAV, Mono, 44100 Hz, 32-bit float

---

Here is macro “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX”:

The macro is similar to Koz’ “Audiobook-Mastering-Macro” but adding Noise Reduction at the start and Low-Pass Filter at the end. I also read through radioeng & Koz’ discussion.

---

Here are my sample WAV files:

NOTE: I ADDED THE TWO BIG BREATHS ON PURPOSE! I may be doing some breath volume reduction or noise gate processing later, and might need your thoughts on that, too.

Raw unprocessed WAV file:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 6,6,6:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 12,6,6:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 18,6,6:

---

QUESTION 1: Should I “Normalize peak amplitude to 0” before using Noise Reduction?

QUESTION 2: I hate the fan noise, but is NR beyond the Beast too much? My goals are ACX acceptance and a good-sounding audiobook.

Thank you, I appreciate your time and assistance.

Zash

---

Audacity 3.5.1
Samson Q2U USB/XLR Dynamic Microphone
EPOS Sennheiser GSP 300 Headset

Windows 10 Pro Version 22H2 10.0.19045 Build 19045
HP EliteBook 840 G3 x64 8GB RAM
Intel Core i7-6600U CPU 2.60GHz 2.81 GHz

---

[Zash]

Well, now the plot thickens (the audio recording plot, not the book plot) <— feeble attempt at humor

My goals are ACX acceptance and a good-sounding audiobook.

The booster fans are part of this building’s A/C system. I was able to turn them off through some creative circuit breaker use (shh). But tonight I encountered the sound of the outside A/C unit for an adjacent building, which I obviously cannot control.

So the chart above shows the Noise Floor for the various room conditions after processing with my Audacity macro described above with different Noise Reduction settings (6, 12 or 18). I’m not sure why the noise is more effectively reduced for the booster fans than for the outside A/C. I recorded the booster fans last night and the other two tonight, but I believe everything was treated the same–except the raw peak level is higher tonight, meaning the microphone was probably closer.

Here is the data:

I would greatly appreciate thoughts on what level of noise reduction is best for my submission to ACX, while at the same time making the audio sound as good as possible. (Some ACX-passing noise levels sound terrible to me.)

I have read in many threads on this forum that all chapters need to sound the same. So I’m concerned if that A/C unit is running for some chapters, not running for others, or coming on in the middle of a chapter. That is one reason the 18 dB Noise Reduction is attractive to me. Or perhaps even more reduction and then add in some room noise I capture with no fans and no A/C unit and then process with appropriate parameters to “satisfy” ACX. I would really appreciate your learned advice here.

In case they would be helpful, here are the WAV files.

---

NOTE: The sample WAV files with the inside booster fans running are in the original post above.

---

SAMPLE WAV FILES, NO FANS, JUST OUTSIDE A/C UNIT

Raw unprocessed WAV file:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 6,6,6:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 12,6,6:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 18,6,6:

---

SAMPLE WAV FILES, JUST ROOM NOISE (NO FANS, NO OUTSIDE A/C UNIT)

Raw unprocessed WAV file:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 6,6,6:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 12,6,6:

Processed with “Noise-LowRolloff-RMS20-LowPass-Limiter-ACX” with NR 18,6,6:

---

Thank you.

Zash

[kozikowski]

@Zash

Some ACX-passing noise levels sound terrible to me.

The ACX Noise specification assumes gentle, spring rain in the trees, shshshsh noise. There is a broken USB noise that doesn’t conform to this specification. That’s the “Baby Screaming on a Jet” noise. Noise specifications have to be much higher for those and/or get rid of the noise at the recording stage.

If the fans pump on and off, they’re not “Noise”. They’re performers in your show. Audacity usually can’t split mixed shows into individual performers.

This is where I would drive to the quiet park and record in my truck/car/lorry.

If you’ve been reading the mail, you know I’m a fan of recording on my phone. iPhone Lossless Voice Memo does a remarkably good job.

@Zash

that all chapters need to sound the same

Yes, chapters have to match. This is the ACX thing about avoiding distractions.

The good news is you don’t have to be a classically good performer or actor. I really like the Sarah Vowell books.

Koz

[kozikowski]

I need to wait for daytime to listen on my grown-up sound system. I generally don’t like noise reductions stiffer than 12.6.6. I can hear the tonal distortions creep up.

As we go.

Koz

[Zash]

Thanks for the input, Koz. I look forward to your reactions after listening on the right system.

[kozikowski]

@kozikowski

If the fans pump on and off, they’re not “Noise”. They’re performers in your show. Audacity usually can’t split mixed shows into individual performers.

A note on that. Noise Reduction works by memorizing the character of constant sound. The pumps turning on and off isn’t constant.

It’s moving/changing sound. It’s a performer.

As we go.

Koz

[Zash]

Yes, I do understand that a changing level of background noise will not be smoothed out using Noise Reduction.

I have a more fundamental question, I think. “Room tone” often refers to video scenes, where the performers are in a setting that has a natural presence or ambiance. Then recorded room tone can be used to smooth out the audio on edits. That makes perfect sense.

Recording an audiobook is different. The room tone where the narration is being performed generally has nothing to do with the story. So the goal is to minimize extraneous noise like traffic, aircraft, HVAC and so forth, and try to eliminate echoes, i.e. have a very low natural noise floor.

Yet absolute silence can be jarring, and ACX has a noise floor range. But some room tones, like fans, are distracting while others, like your forest rain example, are soothing.

It would seem, then, that driving the recorded noise floor down as far as possible without artifacts such as the wine glass metaphor, and then layering soothing room tone across the entire recording would make the most sense.

Perhaps the Noise Reduction algorithms cannot accomplish the reduction without introducing the artifacts.

[kozikowski]

@Zash

a changing level of background noise will not be smoothed out using Noise Reduction.

Worse yet, the correction will give pulsing noise residue making it even harder to remove.

@Zash

the performers are in a setting that has a natural presence or ambiance.

And we go through a lot of effort to make that ambience come out right. There’s nothing like a movie reveal where they play the scene flat without the ambience. Also note that the ambience comes out of six of the seven surround channels. The Center Channel is dialog.

@Zash

try to eliminate echoes, i.e. have a very low natural noise floor.

Try not to conflate noise and echoes. Echoes are not the air conditioner fans or Metro bus. Echoes are the performer’s own voice arriving at the microphone multiple times. Echoes are super difficult to get rid of because you’re asking the correction to remove the performer from himself. Echoes violate a number of quality production rules.

Both…
Zash_Audacity_Sample–no fans only outside AC–NR6-6-6
and
Zash_Audacity_Sample–no fans no outside AC–NR6-6-6

…are passable. The first one has -65.69dB noise. -65dB is my personal, unpublished limit. If you’re worse than that, you must do something. The other has -71.4dB which is more than enough—as long as the noise doesn’t change.

@Zash

ACX has a noise floor range.

ACX has a noise floor limit.

Audacity ACX-Check has two limits. The normal limit of -60dB and a warning against blackness-of-space dead zero. That last one is part of an unusual standard. Audacity > Generate Silence produces that dead silence and is usually the signature of someone who has no idea what they’re doing and is likely to make other audible mistakes. It biases the acceptance officer to look closely for other errors.

@Zash

layering soothing room tone across the entire recording would make the most sense.

Yes. That’s one of the Get Out Of Jail tricks when you used too many Generate Silences and you can’t find them all. I believe Pink Noise is preferred. It’s less piercing. It’s a serious juggling act if your regular noise is close to failing.

@Zash

Yet absolute silence can be jarring

How many audiobooks do you own and/or listen to? Performer submitted MP3’s are required to be at super good compression quality (192-minimum) so ACX can apply their own corrections and product variations without causing problems. MP3 editing always causes problems and it always gets worse.

All of my purchased books drop to dead silence during the voice segments.

Audiobooks are not Radio Theater.

Maybe a little goal management. Our goal is to make ACX happy so they accept the submission, publish, and make you a fortune so you can buy that beach house on the California Coast.

If your work demands a bunch of corrections, patches, and modifications, it’s a good bet you’ll never get there. Each patch causes other errors. Some studio management might be needed.

If you’re close enough now, maybe adding the Kitchen Table Sound Studio. Scroll down.

Too compressed Rejection

I will master and send to ACX again ACX wants all your chapters to match. I did the record early morning when it was really quiet. That leaves room echoes. I can give you the basic ideas very fast. 1/2" PVC Plastic Pipe. 72" by 80" Heavy Furniture Moving Pads. Four PVC Corners. [Screen Shot 2019-05-15 at 4.27.55.png] Cut the pipe into 8 lengths, each 20". Push the pipe and corners together into a square cave. [PVC-Studio2.jpg] Throw furniture moving pad over it to cover top and t…

Too compressed Rejection

Kitchen Table Sound Studio Revision 1 This is the parts list for the studio. Each of those eight plastic pipes is 20 inches long. No special parts or celebrities. They’re half-inch PVC. Four corners, four end-caps and two heavy 80" x 72" furniture moving blankets. [KitchenTableStudio-PartsList-Rev1.jpg] This is the frame construction. No glue. [KitchenTableStudio-Frame-Rev1.jpg] This is the finished studio. The pipes have end caps and the table has a cover, folded over. [Kitch…

The forum link keeps wandering.

Koz

[Zash]

Thank you for the input, Koz. A few questions:

“All of my purchased books drop to dead silence during the voice segments.”

How did those audiobooks get accepted by ACX… I think I read in some forum posts that a noise floor below -80 dB (or maybe it was -90 dB) will get rejected. Or does it just, as you said, make them look closely for other errors?

“Both…
Zash_Audacity_Sample–no fans only outside AC–NR6-6-6
and
Zash_Audacity_Sample–no fans no outside AC–NR6-6-6
…are passable.”

What do you think of the “character” of the room noise in those clips? When I listen on speakers, they sound fine. When I listen on headphones, the noise annoys me. That goes back to your comment about audiobooks dropping to dead silence. Do you feel my NR12 and NR18 clips have noticeable artifacts? My voice is lower frequency, so I don’t hear the wine glass effect, but perhaps you do. My preference would be as low a noise floor as possible with ACX acceptance.

I haven’t noticed a problem with echoes here. I am surrounded by heavy drapes. Please let me know if you hear any echoes in the clips.

I would choose somewhere other than California, but I get your drift!

Thank you.

Zash

[kozikowski]

@Zash

How did those audiobooks get accepted by ACX

ACX does post production after you submit.

@Zash

Do you feel my NR12 and NR18 clips have noticeable artifacts?

They don’t and it’s not wine-glass distortion. Stiff Noise Reduction can give you Sibilance and Essy sounds. But I didn’t hear very much of that, either.

@Zash

I am surrounded by heavy drapes. Please let me know if you hear any echoes in the clips.

@Zash

the Samson Q2U oblique positioned

Between those two, I think you have good personal presence. No bathroom or kitchen sound.

Let’s do one a little stickier. Read some of your own work. Something you’re going to want people to pay for. If that’s too messy, read something similar.

The forum submission limit is 4MB. The upload icon is the thick horizontal bar with up arrow.

Remembering I need to be able to buy your book on Amazon right now for you to submit. That and it should have plot, characters, and setting. No cookbooks.

My favorite variation is The Hero’s Journey author Joseph Campbell interview where they talk about plot, characters, and setting. That counts.

@Zash

I would choose somewhere other than California

Majorca, perhaps?

Koz

[Zash]

Ok, great, I’ll work on a realistic sample.

@kozikowski

ACX does post production after you submit.

I don’t really understand your answer. If ACX gets a submission with a very low noise floor, what does their post production do to the files?

Majorca is getting warmer (<— another feeble attempt at humor) but I’m not a beach person. More like a private island on an inland U.S. lake, with a great house and nice boats. It’s good to dream.

Zash

[kozikowski]

@kozikowski

ACX does post production after you submit.

@Zash

I don’t really understand your answer.

ACX’s job is not post your work for sale and walk away. For one example, my purchased Audiobook CDs have multiple CD tracks per chapter and the tracks all come out exactly even in Audio CD format, have multiple physical CD breaks, and they are all perfectly clear and have velvety quiet background “noise.”

Somebody had to do all that, to manage that product.

I had brief contact with one of the post production team when ACX’s quality control went into the mud and they started turning out audible trash. I want to call that a Covid Event, but I don’t remember.

Somebody has to do the submission acceptance dance when a performer posts chapters for their book. We are reminded that there used to be a whole office of people who did that on demand and all you had to do was submit a one or two minute sound test and they would review it. That’s when I found that although my file making abilities were practically perfect in every way, my voice could use work and I gave up.

Not all purchased Audible Exchange products have convenient chapter breaks and the work has to be shuffled.

And etc.

So all you’re doing is getting your chapters within range of ACX Post Production process.

Now that the short clip audition process is gone, you are urged strongly, by them, to be really, really careful to make your submitted chapters as perfect as possible when you submit The Whole Book.

How to Succeed at Audiobook Production: Part 1

`

And then there’s the forum where we can point out errors, hopefully before you reach the last chapter.

Koz

[Zash]

Thank you for that explanation. I now understand much more about ACX’s role in the process.

The four-part ACX “How to Succeed at Audiobook Production” you linked to is very helpful. I have a question, though. In the “Part 3 - Mastering” video at 2:36, they present this…

---

Do not use the following plugins:

• Noise reduction
• Click/pop removers
• Gating
• De-essers
• Audio effects processors (reverb, chorus, delay, phasing, etc.)

---

and later, this…

---

Only three tools for mastering:

• Equalization
• Normalization
• Compression

---

So… I can see not using click/pop removers (instead punch-copy-paste room tone, or rerecord a portion if needed), gating (edit down breath volume or replace with room tone), or de-essers (use a better mic obliquely).

Using equalization, normalization and compression is essentially the “Audiobook-Mastering-Macro”.

That leaves Noise reduction and Audio effects.

Noise reduction – Are most successful ACX submissions really mastered with no noise reduction? I understand the “best studio and technique” philosophy. But does ACX’s software analysis really pick up that, for example, NR6-6-6 was applied?

Audio effects – At first blush, these effects seem amateurish (types the amateur). Then again, a very subtle effect for a character’s thoughts vs spoken word doesn’t seem so horrible.

Similarly, does anyone ever successfully use quiet music between chapters, beneath the narration of the chapter number/title? It can be annoying, like in Stephen King’s Insomnia, but used very sparingly it might add to the experience.

[Zash]

@kozikowski

they are all perfectly clear and have velvety quiet background “noise.”

What does ACX apply to achieve that velvety quiet background?

I have a related question about audiobook distribution.

‘ACX distributes your audiobook to Audible.com, Amazon.com, and iTunes, the three leading retailers of audiobooks, as well as their global retail partners.’
Audiobooks Through ACX

I have also read this thread:

Audiobook distribution

This really does have something to do with creating files, I just didn’t know what to name the topic. A lot of folks here talk about ACX, and I remember a while back I was helped with producing a book and getting stuff together. I have a distributor for my own stuff that’s NOT ACX and I was looking at the specs for it today as I uploaded. I realized hey, there may be folks here who can use this information. They’re called Author’s Republic - and I’ve always gotten more from them than Amazon. …

I assume that ACX does not give us back their processed files, with the rights to submit them to competitive audiobook sites. Do all of these other sites (Google Audiobooks, Kobo Audiobooks, Downpour, etc) apply processing similar to ACX? Or will some of them simply publish the file as-submitted, assuming it passes their requirements, and if so should processing (e.g. noise floor) have been applied before submitting to them?

[DVDdoug]

@Zash

What does ACX apply to achieve that velvety quiet background?

…I assume that ACX does not give us back their processed files

I didn’t THINK they do any audio processing. But maybe they do convert it to a different format for copy protection. (I don’t believe MP3 can be copy protected.)

The other sites shouldn’t be “enhancing” your audio either. But they may be adjusting the volume. All of the popular music streaming sites and YouTube volume-match files. And most streaming services use their own file-compression format (not dynamic compression.)

[kozikowski]

@Zash

What does ACX apply to achieve that velvety quiet background?

Not a clue. They would just as soon let you assume they don’t do anything to it. And that worked really well right up to the pandemic when they got some ratty production products.

In general, if you can pass ACX and don’t have unusual noises like that screaming mosquitoes thing—and—you read in a theatrically smooth style, you an submit anywhere.

The worst that can happen is you don’t fit in the client’s loudness model, and even then, they may just correct it themselves to get the show out the door. You can change loudness in a lot of programs.

By the way, this is the clear reason your Edit Master should be a WAV file and not MP3. You can’t change or edit an MP3 without adding damage.

You can also make an Audacity Project in addition to the WAV, but Projects can be brittle and not as reliable. Search the forum for “My Project Won’t Open!!”

Koz

[kozikowski]

Noise Reduction of the Beast (6, 6, 6) was chosen because it’s effective and you can’t hear it working.

My last sound test produced a Mastered clip with -61.8dB noise. Technically that does pass, it’s quieter than -60dB, but it’s too close. All you need is someone measuring noise slightly differently from Audacity, and the clip will fail.

I applied 6, 6, 6 and got -68dB noise. Done, and out the door.

ACX doesn’t like you applying mouth noise filters, noise reduction, and advanced processing. This snaps us back to their law of Avoiding Distractions. What’s the possibility that a new performer with no technical knowledge or experience is going to read a ratty chapter and get all the adjustments perfect on a pile of corrections?

There is a sister forum posting where we are weeks into explaining how the Audacity Compressor works. That’s one correction.

There are YouTube shows I can’t watch because of bad sound. I don’t care how engaging it is, it makes my ears hurt. It’s distracting.

Koz

[TSOS]

I have had the same issue. I suggest you do the same thing that I did. And you will be amazed at the results. I can’t see your spectrum View all that well, however, I would put a notch filter of 690 and a q value of 10 and see if that erases that line. Then do the same Notch filters to the other lines, however, the Q filter should be set at a higher volume so is not to take away from the recording. I would set this at maybe 15 or 20 Q value. Then, take the remaining intro sound use that as a noise profile. Have your settings at 12, 12, 0.

[Zash]

TSOS, thank you for your post. The bright line at 700Hz was from the booster fans in my building’s A/C system, and I found a way to turn them off. I’m getting intermittent noise from the outside A/C unit for an adjacent building. I tried recording in my SUV using the Android app Smart Recorder by SmartMob v13.0, but I prefer the sound I get with the Samson Q2U mic in my office. So I’m just recording when that A/C unit isn’t running, which is mostly at night when the world gets quieter anyway.

I did try a notch filter on my breathing noises. The inhaling noise is much wider in the frequency domain than the fan noise, so applying the notch filter across the full time domain recording changed my voice too much. Instead I was looking for a gated notch filter and never found one, but I’m currently happy with gated low-pass filtering using Audacity’s Noise Gate “Gate frequencies above (kHz)”; see my discussion here.

I honestly haven’t experimented much with the Sensitivity and Frequency Smoothing (bands) settings within Noise Reduction. I just now tried 12, 12, 0 and it seemed to mess with my “ess” sounds a bit.

Applying Noise Reduction to one of my chapter files recorded with just room noise (no A/C or booster fans), NR 12,12,0 and NR 12,6,6 have almost identical ACX Check results.

NR 12,6,6
Peak level: -3.50 dB Pass
RMS level: -21.20 dB Pass
Noise floor: -82.18 dB Pass

NR 9,6,6
Peak level: -3.50 dB Pass
RMS level: -21.19 dB Pass
Noise floor: -79.23 dB Pass

NR 6,6,6
Peak level: -3.50 dB Pass
RMS level: -21.18 dB Pass
Noise floor: -76.28 dB Pass

Koz, you corrected my, “ACX has a noise floor range” to “ACX has a noise floor limit” – that is, a noise floor higher than -60dB RMS will be rejected. Also, that dead silence “biases the acceptance officer to look closely for other errors.” Is there any ACX rule-of-thumb as to what is close enough to dead silence to raise that red flag… do you think it’s -80dB? -85dB? -90dB? I ask because I really prefer the NR 12,6,6 result, while in 9,6,6 and 6,6,6 I can still hear the room noise (when using headphones). I’d rather the room noise be as low as possible while still getting ACX acceptance.

Thanks.

Zash

p.s. From this topic, it looks like a noise floor below -90dB is a problem, and perhaps the best range for ACX submission is -65dB to -80dB. Thoughts?

[kozikowski]

We note that all these discussions refer to random electronic or background fan noises. I call them spring rain in the trees FFFFF noises. if you really offended the production angels you can have USB damage which is more like baby screaming on a jet. That kind of noise does not work with the -60dB limit. That noise violates the ACX Distraction rule. No Distractions.

We also note that published audiobooks frequently have nearly zero background noise. This suggests strongly that they do post production after you submit. There are other reasons to believe that. So the fact that you can hear some gentle hissing back there with good quality headphones doesn’t count. It just has to pass the limit standard and not be distracting.

Another note. I can’t prove this one, but I suspect strongly that absolute dead zero background noise is the red flag that the performer, generally an inexperienced production manager, has mis-managed the tools and caused damage to the voice. As you said in your post, you don’t use some corrections because they cause audible damage.

Both Audiobook Mastering Macro and Noise Reduction (6, 6, 6) (I know it’s misleading, but how can you resist “Noise Reduction of the Beast?”) were chosen because they can do their jobs with no audible damage. You sound exactly like you after processing and you usually pass ACX requirements.

There is one slippery Noise Reduction note. The first number can be boosted as far as 10 or 12 without being audible. But that usually means you need to address your studio, not throw extra patches, corrections, effects, and filters.

Did you notice that On-Line ACX Audiolab does not check noise? The one standard home performers are likely to fail.

@Zash

perhaps the best range for ACX submission is -65dB to -80dB.

That should work. I personally assume -65dB as the “real” limit number, not -60dB. This is against post production steps causing additional noisy damage. I think it was my iPhone voice test that actually did pass but with slim margins -61.6dB. So I did the Noise Reduction anyway. My “final” came in at -67.6dB.

-96dB is the quiet limit for any 16-bit sound. There is never anything quieter than -96dB.

Similarly, Audiobook Mastering does not use -3dB as the peak limit. It uses -3.5dB. The specification is no louder than -3dB and post production processing will almost certainly mess with this number.

Koz