Noise floor question

Hi everyone. I have recorded/edited 15 chapters of my 19-chapter book, which I am self-publishing this summer. And while I believe the audio sounds good, I am unsure whether or not it will be accepted by ACX. The reason that I am unsure is that whenever I run it through the “ACX Check” on Audacity, it generally shows that I pass in two areas (Peak Level and RMS level), but I am given a warning in regard to the noise floor. It is like this for almost every chapter that I have recorded. What I’m wondering is, will ACX reject my files because of the noise floor issue? And if so, is there anything that can be done to raise the noise floor in the chapters that I’ve recorded? Any help would be greatly appreciated.

Which message did you get? There are two. Your noise is too loud and your noise is too quiet.

Home users never pass noise. Home microphones intentionally record with low volume and that puts you at the mercy of noisy preamps, audio interfaces, and computers. Hissy “rain” or leaking air riding through everything, or worse, listening to the downstairs neighbor’s dog.

The Low Noise failure is a little more magic. You can’t get a noise floor reading in the -80s or -90s at home without messing with sound filters, special effects, and processing and unless you’re really good at it, that can cause voice distortion. So that’s what ACX Check is looking for.

How did you get your chapters and what is your noise floor?

Probably the worst thing that ACX did in the Age of Covid was kill off their pre-submission testing. Back when I did it, I submitted a two minute voice recording for testing before I read the whole book. They came right back and said I had submitted an absolutely perfect sound file, but my voice pretty much sucked so don’t give up the day job.

I didn’t.

They don’t do that any more. The first time you experience ACX Evaluation is when you submit the whole book. Or You can submit samples here and we can try to catch problems before you submit.

You can only post a bit over ten seconds of perfect quality WAV sound on the forum. We posted a pre-baked sound test for evaluation.

https://www.kozco.com/tech/audacity/TestClip/Record_A_Clip.html

Read that the way you have been reading the finished chapters.

Alternately, do you have any of your finished chapters available on a file sharing service so we can listen to the real thing?

Please note you can’t edit a submitted MP3. You have to patch or edit your WAV Edit Master and then make a new MP3 for submission. This can come as a nasty surprise for people who don’t make WAV edit masters.

Koz

Is the noise too high or too low?

Too low usually means you’ve over-processed (noise reduction or a noise gate) or you added-in a section of pure-digital silence.

If there is too much noise (normal if you don’t record in a soundproof studio) the first thing is to run the low roll-off for speech. The low-frequency noise is usually the worst but you can filter-out the lowest frequencies without affecting the voice.

The [u]Recommended Audiobook Mastering Process[/u] includes low-frequency roll-off, and it has some suggestions if there is remaining noise.

Note that regular-linear volume adjustments change the noise level along with the signal level… Typically RMS Normalize ends-up boosting the volume so it makes your noise floor worse (less negative).

So post back how you’re producing your work and include a sample if possible. We’re on the edge of our seats.

Koz

Hi Koz,

On this note…

I’m also reaching “magic” levels of noise floor. I’ve removed the single post-production plugin I was using (De-Breath) and then ran the Audiobook Mastering Macro. Where many of my chapters pass the noise floor test at around -84, I do get failures of -98, -107, etc. This is doubly strange since I know for a fact that some of these recordings passed all 3 ACX Check parameters only a few weeks ago but in compiling this post I’ve gone back and they are now failing at passing noise floor.

Is the ACX Check plugin using other fluctuating data when crunching these numbers?

How can I be achieving such low periods of noise floor when I neither recorded with processing nor am currently running processing in my effects chain?

Scratching my head pretty bad on this one. Appreciate your time,

Jamie

Abandon your process for about ten seconds and read this.

https://www.kozco.com/tech/audacity/TestClip/Record_A_Clip.html

It is possible to get ACX to evaluate your work with their ACX Audiolab service.

https://www.acx.com/audiolab

Fair warning before you go charging over there, they don’t check theatrical presentation (human quality control) and they don’t check noise. In other words they check the same things ACX Check does, maybe slightly differently. Whatever you can do without the humans that called in sick.

ACX Check does have an oddity. It rips through your chapter and finds the lowest noise reading … and then reports that. If anywhere in the chapter the noise drops to dead black zero, like -90dB, however briefly, that’s the number you get.

Check your chapter after you get done editing and correcting the mistakes but before you apply any effects or filters. For example if DeBreath reduces your quick breath sound to -90dB, then that’s where the measurement is coming from.

Fair warning, removing breath sounds and gasps is not supposed to leave a hole. It’s supposed to dip to your normal room background sound so as to sound natural. That’s my problem. By the time I get done correctly filtering out all my mouth sounds [tick][gasp], I’ll be in a old folks home.


Two audiobook questions: Can I buy your book on Amazon right now? That’s required for publication.

Does your book have Plot, Characters, and Setting, or does it talk about Plot, Characters, and Setting? Those work best. You can’t publish a cookbook.

Scroll down.

https://www.acx.com/help/200878270

One of my favorite publications is Hero’s Journey by Joseph Campbell where he talks about plot, characters, and setting.

Koz

Hi Koz,

The book is an original novel. Exciting that it’s almost out there!!

But these technical hurdles are very frustrating…

Firstly, I just checked out that ACX Audio Lab analyzer you mentioned in the previous post. Unfortunately, the “Audio Lab does not detect issues such as noise floor or editing errors, so remember to conduct your own thorough review of your files before submitting finished audiobooks to ACX.” The only way I know how to do this otherwise is by using your ACX Check plugin on Audacity which you have mentioned only detects issues w/o giving more exacting detail. I’m a cyclist and I was thinking about when I get a flat tire in the middle of nowhere. As a cyclist it’s imperative for me to find the exact location of a puncture(s) on my tire in order for me to be able to patch the hole(s). How then can I identify the location of potentially one infinitesimally small moment of “dead silence” in my recording? Is there a meter or other visual indicator (like the red spikes when distorting) that I can implement either automatically or by scanning the entire recording in real-time using a plugin? Again, with this particular recording I had not used any processing–during the recording or in post–and am only using the Audiobook Mastering Macro in Audacity immediately before exporting the finished file for submission.

And now I’ve encountered a new issue with regards to the Audio Lab…

The Audio Lab is apparently measuring my RMS (and possibly other data) differently than the ACX Check plugin. This same recording was measured by the ACX Check plugin as -21.18 dB (Pass). The measurement was taken immediately before exporting the file to submit to the Audio Lab. When I then had the Lab measure the same recording I received a failing measurement of -25. Now I do understand that RMS can be measured in different ways by different tools. But how do I know exactly how ACX is measuring RMS and how I can then recalibrate Logic Pro X and/or Audacity plugins to measure for RMS in this exact same manner?

Could there be something in the exporting/encoding process that’s manipulating the structure of the recording?

Down the rabbit hole I go

Thanks, Koz

Do you have WAV copies of all your raw readings? No filters, no effects, no corrections. Just as you read them. That’s highly recommended. Keep it/them in safe place, divorced from all your editing and production. It’s a New User error to announce directly into an edit session and then on to the finished production. If Audacity or the computer goes into the bin anywhere in the process, it can take the show and all the work with it sending you straight back to announcing into a microphone in a quiet room.

ACX requires submission as MP3 compressed sound file chapters with at least 192KB quality. You need to know that if you make a mistake in the submission, you can’t correct the MP3, reexport, and resubmit. Every time you make a new MP3 from an old one, the quality goes down. The new one isn’t 192 any more. You have to edit your perfect quality WAV Edit Master and make a new MP3. That’s why a WAV Edit Master is a terrific idea.

Audio Lab does not detect issues such as noise floor or editing errors

Correct. I think I mentioned that. They check everything a machine can check. Noise is hard.


ACX uses RMS - Root Mean Square as the loudness standard. It’s a math function. It’s the area under the curve or the amount of “energy” in the chapter. I don’t know of different ways to detect RMS.

Some production companies use LUFS instead of RMS. That one can determine loudness, too, but it takes into consideration the odd things your ear does when it hears things. For one example, your ear doesn’t work very well in very high and very low pitch sounds. LUFS knows this. RMS doesn’t.

A word on how Audiobook Mastering works. There are some tricks to it. We are cautioned sternly to apply the three Mastering tools one after the other (or apply the one-step Macro version. Same thing). Do not add any tools or leave any out.

Filter Curve, the first tool is a rumble filter. It gets rid of thunder, earthquakes, heavy trucks going by, and some thumping breath and wind noises . It pre-conditions the chapter for the next tool, Loudness Normalization. That’s RMS. There was a recent poster who set that for LUFS by accident instead of RMS and started getting odd chapters. Yes. You would. RMS Normalization cranks the overall volume of the whole chapter up and down until the “energy” or loudness is perfect. The chapter goes on to the Soft Limiter which gently folds the tips, peaks, and high points over until they pass the ACX peak specification.

I bet you’re wondering to yourself why doesn’t the limiter affect RMS. It does, but not much. Compared to the massive, ground-shaking work RMS is going, changing a light, delicate tip or peak here and there is a piffle.

Please note that the chapter before Mastering and after is significantly different. “After” doesn’t have thunder and earthquakes. Those two can have very serious energy issues as anybody that has every been through an earthquake can tell you. Weirded yet, you can’t always hear them.

That’s why a similar rumble filter is found on most outside broadcast sound mixers, field sound mixers, or recorders.

And that’s why if you submitted before and after mastering to ACX for testing, it may be different. ACX Check will be too.

Koz

Home microphones can get their hand in, too. It’s not unusual for home microphones to make their own rumble and thunder. The manufacturers don’t pay any attention to it because it can be expensive to fix and nobody can hear it anyway.

Koz

You have to edit your perfect quality WAV Edit Master and make a new MP3. That’s why a WAV Edit Master is a terrific idea.

You can’t make your WAV Edit Master from the submitted MP3. You have to make it from the edit. A WAV from an MP3 isn’t “clean,” It’s got the original MP3 damage burned in and you can’t stop it.

MP3 can be a land-mine time-bomb

Koz

Hi Koz,

This is Jamie, not Chris. I tacked my Noise Floor questions to the original post.

My issues are all post-Mastering. I am not a new user and I have all my original recordings stored on Logic Pro X. I export them as WAV to Audacity where I have been using the Audiobook Mastering Macro. I then use ACX Check. And then I export to MP3 to submit to the ACX Audio Lab. During those last two steps is where I have been receiving different readings.

ACX themselves say that RMS can be measured in different ways which has people/machines arriving at different results. I excerpt their warning here:

"RMS values can be measured using multiple methods, so readings can differ depending on your DAW and the plugin/meter used. "

That must explain why the Audacity ACX Check plugin reports a passing RMS immediately before the Audio Labs analyzer reports a failing RMS measurement.

It turns out that in this specific instance I had enough headroom and was able to boost the gain of the entire recording back in Logic Pro X in order to then run the resulting WAV file through the Audacity Audiobook Mastering Macro which produced a higher RMS reading that passed both the ACX Check plugin and the Audio Labs audio analyzer. But the discrepancy between both tools’ readings is still less than clear.

And to go back to noise floor…

I understand that the low frequencies are rolled off in Mastering. I still don’t know how to eyeball when “dead silence” occurs. Other resources online prescribe boosting the low-end frequencies to raise the noise floor enough to pass inspection. This didn’t immediately strike me as a great idea for many reasons. And I would still like to be able to pinpoint impossibly low instances of noise floor in order to perhaps more manually address such an issue(s).

I think the forum folded your new post over into a new chapter (2) and you didn’t see to follow it.
I also patch up the text in your posts.

This is also why it’s a shaky idea to post onto the end of someone else’s work. Suddenly your post isn’t where you thought it was.


I don’t know a good way to sort where a chapter is hitting dead-black-zero. There are no handy tools for that like the red spikes when something is likely to exceed 100%.

I can think of a desperation method. Select half of a chapter and apply ACX Check. If it has bad room tone, divide it in half again. Keep checking. My bet is you’re going to find silent snippets sprinkled all around the chapter because that’s how your recording system created it.


However, your microphone should not be doing that, so it or the system is “broken.” It would be good to find out why. The most likely candidate is multiple sound apps. Skype, Zoom, Meetings and others create a supervisory environment and you can’t stop them. Sometimes even when you close them, they leave their tools and settings running in the background. They run their own noise reduction, environment management, echo suppression and noise gate. That’s how they get away with having six people on a screen checkerboard and still be able to hear everybody clearly. No meeting is going to have all six people announcing in a quiet studio.

I’d be tempted to shut down the machine (not restart) and don’t let it volunteer to open everything up again when you start. Just open your recorder software later and see it behaves OK. Keep peeling off apps and kit until you get a stable, predictable results.

Are you recording in Audacity? Audacity doesn’t apply any corrections, filters, or effects or tonal management during recording. Other applications may have settings to Automatically “Clean Up” during the performance. It would be good to know that.

ACX doesn’t like your system doing that dip to dead black thing because they play games with the chapters themselves and you could get conflicts and funny sounding work.

Yes, I saw a Youtuber describing an equalization technique where he boosted extreme high and low pitched sounds (V-shaped equalizer) until the background noise came up just enough to pass without being able to hear it in the show. That makes my teeth hurt, but that does work if ACX accepts it. I think I saved that link here somewhere.

Koz

I think that’s it.

https://youtu.be/7ihmSXcQ3i0?t=301

Koz

Since you’re on a Mac, you should know that the Mac has tools and corrections. Apple (upper left > System Preferences > Sound > Input.

The built-in Microphone has Ambient Noise Reduction.

Screen Shot 2022-06-28 at 17.47.45.png
I’ve never used noise reduction, but I do know that if you’re careful, you can do some fine sound recording. The older Macbook Air has two microphones and can do directional recording.

Koz

Thank you, sorry about the “folding over” multiple posts thing.

I do use a Macbook Pro. I record using Logic Pro X with an external NT1-A Rode mic that’s hooked up to a Scarlett Solo interface. That’s a serviceable idea to run the ACX Check plugin on only parts of a recording to see if the “dead silence” can be isolated. You may be right that Zoom or something has automatically been processing sound on the internal sound card. Hmmm… Since I recorded the book over many recording sessions 80% have escaped whatever’s going on. It may be that the 20% of my book was recorded when I forgot to close Chrome or Zoom or something. I suppose those affected tracks will need to have all their gaps manually plugged with room tone. But I’ll go through the problem tracks using the ACX Check plugin selectively to see if I can indeed isolate the few (or, yikes, many) instances of “dead silence” being measured.

Thanks, Koz

One of the ‘ACX Things’ is your chapters have to match.

There was a forum posting from someone who, in the middle of a book, needed to move houses. I would have started over, but they figured out a way to make a similar enough sounding studio and power on.

I’m betting you’re going to find that patching existing chapters is so amazingly painful that you just make fresh coffee and read them over.

Koz

You could try creating a new track in your chapter and fill it with generated noise or from a repeated sample of room tone in your recording, and then amplify it down -70dbFS or so. This will fill in any dead silent gaps with a bit of noise. Yes, this will very slightly increase your noise floor, but unless your existing noise floor is close to the edge of the maximum allowed noise level (-60dBFS, I believe) it shouldn’t be a problem.