rms normalization relation to limiting help?

My audiobook narration didn’t meet values on ACX, I downloaded ACX Check and RMS Normalize Plug Ins and have been trying to get the settings right for an hour and a half. I stumbled on to them for 2 chapters…but have 23 to go and have figured out the same settings won’t work every time. Can someone explain the relation between the peak level, rms level, normalizing, and limiting so I can do more than stab in the dark? I know there has to be an algorithm, but if I adjust one, it messes up the other. I really would appreciate ANY help here. I am using the latest version of Audacity and am using Windows 8.1. Thanks


This is a long-winded way of introducing a three-tool process. It is a “suite,” a harmonious grouping. The tools create intentional errors and clean up after each other. Don’t ad-lib.

If you have a quiet, well-behaved recording, it’s possible to go from Mastering 4 directly to making the MP3 for ACX.

If it still fails, you can announce a test sample and post it on the forum.


Home recording artists rarely pass noise the first couple of tries. It’s hard to describe how quiet -60dB is. In English, they want your background noise to be a thousand times quieter than your voice.

And while I’m here, it’s good computer hygiene to make WAV (Microsoft) 16-bit protection sound files of your original readings and your Edit Masters. Audacity Projects are not recommended for this.


What do you mean by audacity projects not recommended for this?

Audacity saves Projects but exports sound files. Projects are used when you want to save your editing environment and everything in it except UNDO. They are, by their nature, complex and more brittle than single, stand-alone, perfect-quality sound files such as WAV (Microsoft) 16-bit.

If the machine or Audacity crashes or experiences some difficulty while editing or during saving, it could destroy the show. If you have a WAV backup of the original reading or performance, you can open that up and keep right on editing. If you have no WAV backup, you go find the book and read it again.

Same with the Edit Master. You can’t edit an MP3 without additional sound damage, but you can edit an exported WAV of the final show.

There is a variation. You can Save several Projects each under different names as you go. That does work, but WAVs will open in all three computer types. Projects only open in Audacity.


The “level” of an audio signal is measured relative to “full scale” (also called “0 dB”). In Audacity, “full scale” is the height of an audio track, and is shown on the track’s vertical ruler as +1 to -1. It is measured as “the distance from the (silence) centre line”.

“Peak level” is the absolute maximum distance from the centre line of the waveform. Thus if the waveform peaks at +0.5 on the vertical scale, or peaks downward to -0.5, then the peak level is “0.5”.

There are two common ways that levels are measured in Audacity:

  1. “Linear scale”. This is the scale on the vertical track ruler. It has a “valid” range of +1 to -1.
  2. “dB” (decibel). This scale represents a “logarithmic ratio” of the “level relative to 0 dB”. If the waveform has a peak of +1.0 (or -1.0), then the level is “0 dB”.

Conversion between “linear scale” and “dB scale” can be a bit confusing, so here are some common conversions (close approximations):

1.0 linear = 0 dB
0.5 linear (1/2) = -6dB
0.25 (1/4) = -12 dB
0.125 (1/8) = -18 dB
0.0625 (1/16) = -24 dB
0.0 linear = -infinity dB

Note that the dB values are negative. That’s because the reference level is defined as 0 dB, and valid signals are always less than this reference level.

Note that halving the linear amplitude is a decrease of -6 dB. Similarly, doubling the linear level is an increase of 6 dB.

RMS (“root mean square”) is a kind of average level. RMS is usually measured in “dB”, which is very convenient as this relates directly to dB as used in peak level measurements. Changing the RMS level by 1 dB will cause the peak level to change by exactly 1 dB.

If you have a waveform with an RMS level of -20 dB and a peak level of -4 dB, and you amplify it by +4 dB, the resulting waveform will have an RMS level of -16 dB (-20 + 4 = -16), and a peak level of 0 dB (-4 + 4 = 0).


A “Limiter” effect is a kind of “dynamic compression”. It “compresses” (squashes down) peaks in the waveform so that the waveform has a more even level.
Obviously, by squashing the peaks, the “peak level” is reduced, but there is only a small change to the “average” RMS level because it is only the highest peaks that are affected - most of the audio will normally remain unaffected by limiting.

Example, using Limiting and RMS normalizing together.

Say that you have a recording with the following measurements:
Peak level = 0 dB
RMS level = -25 dB

and you want the audio to have:
Peak level = -3dB
RMS level = -26 dB

Clearly we want to reduce the peak level by -3 dB, but we only want to reduce the RMS level by -1 dB.
From the above, we can see that amplifying (or normalizing) won’t do what we want - amplifying by -3 dB will give us the required “peak” level, but the RMS level will go down to -28 dB (-25 - 3 = -28), which is lower than what we want.

How to do this:

  1. RMS normalize to -26 dB. This reduces the level by -1 dB, so the RMS level is now -26 dB (-25 - 1 = -26). It also reduces the peak level from 0 dB to -1 dB.

  2. Apply a “limiter” to squash down all peaks that are higher than -3 dB. This will reduce the peak level to -3 dB, but because it only acts on the peaks, there is very little effect on the average level. Measuring the RMS level after limiting may show a small reduction in average level, but it is likely to be close enough to not matter.

Had enough?

That will only get you past the ACX technical requirements. ACX Check is an Audacity approximation of the ACX tests you have to pass before your performance goes to the ACX Human Quality Control. That’s where you go to die if you can’t read out loud.

You might consider posting one of these tests on the forum. We’re not interested in ripping you to pieces, but we can sometimes identify problems and easy solutions not immediately obvious.