RMS & Peaks

[jamiehjung]

Hello Audacity Team!

First time posting. I stopped working with ACX because Logic Pro X was providing me incorrect (or different) information as to my audio’s RMS and peak levels so that my last ACX job was a nightmare. I had to keep nudging the audio around hoping it would pass QC. My past work had always passed and was sold on the marketplace, but this one particular book suddenly had audio issues. These issues were that the RMS and peak levels did not meet the requirements of ACX EVEN THOUGH when I used the Logic Pro X 10.5.1 Multimeter these measurements were clearly within the restrictions. When consulting with ACX, I learned that audio can be read differently depending on what software is doing the reading. I understand that this could be possible, but then again how could I (or anyone) ever be absolutely sure that their audio is being recorded properly and/or processed properly for any end user?

That’s how I came to discover Audacity Team and your open-source plugins, primarily ACX Check. What a god send, I thought. Now I just have to learn how to create a chain (or macro) of plugins and my final process will be to run the ACX Check before submitting my mixed and mastered audio. Well, I have been having trouble there too. I still have to nudge audio around to fit it within its limits and so many tools that I’m used to as a Logic user are not available or are more time consuming. I have now been trying to record and process audio in Logic using EQ, peak compression, RMS compression, etc and importing those bounced files into Audacity to use ACX Check to read the files more succinctly.

My latest problem is that Audacity appears to read my processed files bounced from Logic differently. I’m back to where I started, I think. Where Logic’s multimeter tells me that the attached file peaks at -2.8 and that the track has an RMS of 19.5 Audacity reads the same track as having a park of -0.1 and an RMS of -26.7. Either every software really does read audio data differently which in that case my question is: How can I always know that my completed audio files will be accepted by ACX? Or the problem is that I am not using these DAWs’ plugins correctly to which my question is then: How do you read a DAWs’ plugins to get a numeric value that I’m possibly misreading?

I have attached this test file to the post so that it can be analyzed. I’m really at a loss here, guys. I’ve read many articles on loudness, metering, compression and watched many tutorials on the subject of mixing/mastering for ACX delivery but I would really appreciate your help on this specific problem I’m having either with these concepts or with misreading data.

Best,

Jamie
MacOS Catalina 10.15.5
Audacity 2.4.2

[steve]

There are two commonly used types of “peak” measurement.

The most commonly used is the absolute highest (dB) sample value. This is what Audacity’s tools measure. I believe this is also what ACX use.

The other is called “Real Peak”. This is usually a small fraction of 1 dB higher than the peak sample value, but in my opinion the difference is usually so small that it’s not worth bothering about. In extreme cases it can be significantly higher, but this is generally only with artificial test signal rather than real-world audio.

It’s also worth noting that because MP3 format is an inexact approximation of the original sound, the peak level of an MP3 may be a bit higher than the peak level of the original.

Taking both of these factors into account, it is wise to allow about 0.5 dB or so to be sure that your peak does not exceed -3 dB. If your peak level in Audacity measures -3.5 dB, then you can be pretty sure that the MP3 will peak below -3 dB.

The peak level of your audio sample is -0.1 dB.

— I’ll write about RMS in a separate post —

[steve]

There are many ways to measure RMS, but in all cases they are a kind of “average level”.

Audacity’s tools measure “Z-weighted” (also known as “unweighted” or “zero weighted”) RMS. ACX specifications say that they also use this method.

Other forms include A, B, C and D weighted (https://en.wikipedia.org/wiki/A-weighting).
In each case, the RMS is measured after applying specific filters to the sound.


In the case of stereo signals, there are several ways to measure the “average”, but they boil down to two methods: the right way, and other ways Audacity does it the right way. Technically “the right way” could be described as “the square root of the mean of the square value of every samples in both channels”.


Because RMS is a kind of average, long periods of silence within a track will reduce the average level. From a discussion with one of ACX’s technical staff we learned that when they measure the RMS, they ignore (remove) any long periods of silence from the recording before measuring the RMS level.
Taking your audio sample as an example, the RMS (zero weighted) of the entire sample is -25.36 dB, but if we just select the actual audio (as shown below), the RMS measures -21.97 dB.

When measuring a long track (say 30 minutes), a few seconds of silence here and there will not make much difference to the RMS level, but again it is wise to aim for a level with a reasonable safety margin - ensure that your RMS level is well within the range that ACX specify.

[DVDdoug]

We also have a [u]Recommended Audiobook Mastering Process[/u] which will perfectly-nail your RMS & peak levels.

The RMS Normalization adjusts the level up-or down to hit the RMS target. Following that, the peaks are almost always too high so the limiting pushes-down the peaks with very little effect on the RMS levels.

Your noise levels are “artificially low” and you might get rejected for too much processing/noise reduction. (Audacity’s ACX-check plug-in will pass.)

Where Logic’s multimeter tells me that the attached file peaks at -2.8 and that the track has an RMS of 19.5 Audacity reads the same track as having a park of -0.1 and an RMS of -26.7. Either every software really does read audio data differently.

There should be no question about the peak levels. It’s just the biggest number in the file converted to dB. (Although MP3 compression will sometimes change some peaks so your submitted file might be a slightly different if you don’t re-check it after converting to MP3.)

I believe RMS is calculated in segments/sections and then a average (or maybe another RMS) is calculated to get an overall RMS value. That’s probably because the sum of millions of squared values is to big for the computer to handle. Different audio software uses different-length segments and I don’t think anybody knows exactly how ACX does it. Some algorithms may ignore silent parts. Your little attached sample is about half-silence and when I trimmed off the silence it passes RMS at -22dB.

[steve]

A simple way to get the peak and RMS levels right, is to first adjust the RMS level to around the middle of the required range, then check the peak level.

If the peak level is too high (it usually is), then you can reduce the overall level a bit so that the peak level is closer, but ensure that the RMS is still well within the specified range.

Usually the peak level will still be a bit high, but by now it should only be occasional peaks that are a little over. This can be fixed by using a “soft limiter” (https://manual.audacityteam.org/man/limiter.html).


Note that both the peak and RMS should be reasonably consistent throughout the audiobook. If there are noticeable jumps in level, ACX will probably reject it.

[kozikowski]

We designed Audiobook Mastering to as much as possible reflect ACX’s own published standards.

https://wiki.audacityteam.org/wiki/Audiobook_Mastering

You are urged, strongly, to take the steps in order, not add any, or leave any out. The first step, Filter Curve, is designed to remove the most common home microphone distortions. The other two guarantee RMS (loudness) and Peak and if they’re not needed, they don’t do anything. There are no decisions.

If you record well in a quiet, echo-free room, that may be all you need for submission.

ACX provides a testing process similar to the Audacity ACX-Check, so you don’t have to guess at it.

https://www.acx.com/audiolab

https://wiki.audacityteam.org/wiki/Nyquist_Analyze_Plug-ins#ACX_Check


All that said, home readers never pass noise. -60dB in English means your background noise has to be 1000 times quieter than your voice, and in real life, has to be be even quieter than that. Very few homes, houses, or apartments can reach that without extra work. I know the ad for your microphone insisted you can set up on the kitchen table and crank out audiobooks, but most people can’t.

We also note that this circus will only get you past technical testing. Audiobook submissions also go through rights inspection and theatrical evaluation. Can you document who has publication rights to the book you’re reading? Do you like to spit on the microphone when you talk?

The latest zinger is the need to have your book published on Amazon in either eBook or paper before you submit. That peels off many first-time reader/authors.

“Testing one two three,” is nice, but there is a format for forum submission.

Please don’t ad-lib and don’t apply any corrections at all except cutting to length. Read down the blue links. They’re very short.

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

Koz

[jamiehjung]

Thank you steve and everyone else who’s chimed in! Very useful info about how RMS is measured and taken into account differently by different meters’ filters/algorithms.

Koz: I have been going through the mastering steps offered by your team. I have been experimenting with other solutions as well and did deviate from your prescription with the addition of effects/plugins. Below I have attached the requested voice test. It has no effects/plugins processing its audio and will hopefully serve to give an accurate reading of my home studio’s noise floor as well as my microphone + interface setup.

At the same time as I’m trying to learn how to create a template/chain/macro for ACX submissions I’m also trying to make sure my audio is of a competitive quality when submitting auditions on platforms such as Voices.com, Voice123, etc. I know that that’s not the scope of this post, but maybe it’s a mindset I should abandon. Perhaps I’m treating all vocal recordings as equal–each requiring similar standards and a high degree of processing–when in reality every single recording must be hand tailored to deliver to its specific platform and end user. This could make sense as ACX appears to be so extremely regimented that mastering guides have been created to streamline an engineer’s workflow to flow within the confines (and only within the confines) of a series of effects/plugins for the very reason that deviating from a prescribed workflow can get most people into trouble. Am I starting to get on track with this concept of audio production?

Best,

Jamie

[kozikowski]

Voices.com, Voice123 … maybe it’s a mindset I should abandon.

Not at all. ACX’s specifications are very close to broadcast proof of performance. If you can make it through ACX’s testing, you can post anywhere.

It has no effects/plugins processing its audio and will hopefully serve to give an accurate reading of my home studio’s noise floor as well as my microphone + interface setup.

I applied simple mastering to your clip and got this.

You could submit that. It easily makes it past all three ACX specifications. The voice quality is a little bright and “essy”, but that’s relatively easy to fix.

I’m much more concerned with the background noise or room tone. There isn’t any. Something is processing your performance. I was expecting a gentle intake of air just before the first word, and there isn’t one. The file starts from absolute, total, dead quiet with the first word.

The submitted file (before mastering) has room tone (first two seconds) at -96dB. That’s impossible. It’s too quiet. That’s the absolute limit of the sound system with nothing connected. No microphone, no interface.

So that’s sound processing. Sometimes they will accept that, but sometimes not. What in your system could be doing that?

Koz

[jamiehjung]

Hi Koz,

That is wild. Is it possible my mic or interface are automatically boosting signal suppressing noise?

I am in a very sound-treated closet with a fake door filled with insulation. I’m using a Rode NT1A microphone which is hooked up to my MacBook Pro via the Focusrite Scarlett Solo (2nd Gen) USB interface. I’m using Audacity 2.4.2.

And to confirm I did not add any effects/plugins or in any way process the voice test I submitted earlier today. I’m scratching my head over here too

Best,

Jamie

[kozikowski]

If you were on Windows, there’s quite a list of places to check for automatic sound processing, but not so much on a Mac.

What else do you do or what other services do you use? Mac people are famous for leaving things running in the background instead of actually closing them (I, of course, never do that).

Games, Chat, Zoom, Skype, Other sound programs?

Apple > Shut Down…

Not Restart. Watch the screen and see how long the machine takes to halt. Do you get a spinning daisy for some of the shutdown?

Does it refuse to shutdown?!!? I’ve had a licensing program assume it had a better priority than I did.

Do you use iCloud?

Koz

[jamiehjung]

Hi Koz,

I only have Safari, Pages, and Avast open when I’m recording audio. Otherwise my fans have the potential to kick in and ruin my recordings. I do use iCloud for syncing contacts but not audio or anything heavy or complex. When I hit Apple > Shut Down the machine shut down very quickly. My computer is a brand new MacBook Pro (15", 2019) running MacOS Catalina 10.15.5

Maybe I should call Apple and see if there is an audio specialist who can identify the problem.

Another option might just be to make more noise! I’m a voice & speech instructor so I have excellent breath control. I can let that slip a little. Also, if I keep my closet door open a few inches I will definitely get more room tone and even a little ambient noise through my bedroom window. That seems like a strange fix but if it helps then I’m willing to try it!

Jamie