Normalizing volume across files

Apologies in advance if this has been answered elsewhere - I searched but wasn’t able to find exactly what I was looking for.

I’m in the process of mastering and submitting my first project on ACX. By using the mastering chain suggested in an ACX blog post, I’ve been able to get each of my chapters (files) sounding “good,” but I’m having a hard time getting their volumes to be consistent (there is a noticeable difference in volume from one chapter to the next). I’m sure there’s a quick fix for this – I just don’t know what it is! Please help :slight_smile:

If it’s helpful, my current mastering chain is as follows:

High-pass filter - 48db per 8va, remove 80Hz and lower.
Low-pass filter - 48db per 8va, remove 16KHz and higher.
Normalize peaks to -6dB
Equalize to 1KHz
Limit - max output -3.2dB


We have a handy tool you might be interested in.

ACX has a “robot” or automated quality control system that listens to your work for basic technical standards before a human gets involved. We generated a tool similar to how the robot works called ACX Check. ACX Check can save you a lot of ping-ponging back and forth with them.


Unzip it to acx-check.ny and drag it over to the Audacity plugins folder. Restart Audacity.

Open one of your chapters or segments and select the whole thing by clicking just above MUTE.

Analyze > ACX Check.

You will get an analysis panel similar to attached.

It’s possible you have more problems than simple volume across chapters. For example, it’s rare for first time posters to meet the ACX noise specification.

Screen Shot 2015-08-17 at 19.32.40.png (2.7 KB)

Look at you! You’re two steps ahead of me. Ignore the question that I asked in the Noise reduction thread; you’ve already answered it.

Hmmm – I wonder what the issue is. I attempted to move the file you attached to Audacity’s plugins folder, and each time, I got the error message “Not implemented.” I even tried restarting my computer (because that’s the extent of my troubleshooting skills) to no avail. Any ideas?

Got it to download - when I look in the plugins file, it’s there - but it doesn’t show up in my Analyze menu after I restart Audacity… Is it possible you sent me the Mac OS version instead of Windows, or does that not matter?

It should be generic across platforms. Which Audacity are you using?


The file Koz posted is a compressed-format called a zip-file.
You have to unpack the zip-file : inside are the two versions of the ACX check plugin.
I’ve attached the Windows version …
acx-check.ny (5.6 KB)
If you are using Audacity 2.1.X you will have to enable that plugin, via manage at the top of the effects list,
before it will appear in the analyze menu.

[Correction : the “manage” thing was introduced in 2.1.1 , not in 2.1.0 ]

The latest version of the acx-check plugin is available on the Audacity wiki:
The “download” link will get you a zip file (much like what Koz posted), but in most browsers you can also right-click (or Ctrl-click if your on mac) the “view” link and select “save link as” to save the uncompressed .ny file.

Thank you, Flynwill! I’m really just baffled. I unzipped it, I placed it in the proper folder – I even sent the plugin to a friend and had him go through the exact same process I did (he’s also running Windows 7 Home Premium, but he’s got Audacity 2.0) and it worked perfectly for him. When I open Audacity (2.1.0), there’s no “Manage” option in my Effects menu (or any of my other menus).

The plugin will show up as “ACX Check” in the “Analyze” drop down.

Re-reading your original query would have expected that barring any major changes in your reading style the “Normalize peaks to -6dB” step should pretty much even out volume differences. What I would suggest is to use “ACX Check” and note the “Status A RMS” value for each of your tracks and then use the Amplify effect to change any tracks that are particularly high or low to match the average – you don’t need to be exact +/- 3 dB should be fine.

I just checked , the “manage” thing for enabling effects was introduced in 2.1.1 … Audacity ® | Free, open source, cross-platform audio software for multi-track recording and editing. . I thought it had been introduced in 2.1.0 , sorry :blush:

My posts are all getting crossed - apologies but here goes I’ll try again - if I do this and normalise down it puts my RMS values out - if I then go to amplify and get the rms in limits the peak goes out. In fact on both the amplify and normalise boxes there is a secondary box for ‘max amplitude’ or ‘new peak amplitude’ which varies as the main parameter is changed - I can’t seem to win!

Indeed you’ve still got two threads going on the same subject, but I’ll let the moderators sort that issue out.

As Koz said in the other thread, if the RMS level is too low after normalizing the peak level to ~-3 dB then you’ll probably need to use a compression tool. What a compressor does is to dynamically change the volume, turning it up during the soft bits, and turning it back down during the louder bits. If you over-use it then you may start to notice noise in the quieter sections that goes away when you speak. That “noise pumping” effect can be very annoying, so there’s a limit to what the compressor can do for you.

So you may want to go back to the original performance and try to understand why the dynamic range is so much greater than what is consider typical and see if you can fix it. Do you get much louder when excited (a common problem), it takes practice to sound excited without actually getting a lot louder. Or perhaps you are moving around with respect to the microphone, there’s a big difference between speaking directly into it from 6-8" away and speaking off in some other direction a couple of feet away.

Also as Koz recommend, please post a test of your raw recording.


Hi I’m able to get all my tracks in ACX limits now but would like to refer back to the OP. In an earlier post you talked about the ACX giving a ‘status A RMS’ value but it doesn’t actually do that - it gives an RMS Level and then a separate value for ‘A-weighted’ - is this the one you mean?
I have seven tracks recorded from one of my novels and the RMS levels are 22, 22.5, 21.8,22.4, 22.4,21.3,22.3. - a variance of 1.2db
You said that within + or- 3db would be alright but the ACX parameters are 18 to 23 which is less than that unless you are referring to the A-weighted number which I don’t think ACX refer to in their standards on their web site.
So the question is what parameter in the ACX check in Audacity is the one used to provide consistent volume of tracks and what is an acceptable variation?
I don’t want to do the whole book and then find I need to restart!!!

Yes that’s what I meant, please forgive the “senior moment”. “Status A” is a term use in measurement of photographic film, something I did a lot in past careers.

That’s about a good as you could ask for.

Yes, not sure why they choose what’s effectively +/- 2.5 dB, most of the engineering world uses +/- 3dB as “good enough” but the difference is pretty trivial. Unless you got some intermittent very low or very high frequency noise issue I would not expect to see any significant difference in the variance of the A-weighted numbers and the unweighted numbers.

As far as we can tell (the ACX folks never say one way or the other) the number to look at is the un-weighted RMS number. That’s the one the plugin compares to the ACX -18 to -23 limits.

Brilliant reply and thanks for that - I don’t seem to have wasted any time talking to me an my microphone!!

OK I believe I remember Koz saying somewhere (in a different thread) that it’s good to hover around -6dB, in terms of average volume (I’m sure that’s the wrong technical term, but I think you know what I mean - the little needle should wiggle around the 6 most of the time). When I normalize peaks to -6dB, the “average” goes down to around -20. Is this what’s supposed to be happening?? The only metric I’m still baffled by is RMS - is that referring to that average volume (ideally between -18 and -23)? I know it means root mean square and I’ve had an audio engineer explain it to me, but I still don’t understand how I can know if I’m within the proper range (again, because that darn plug-in just refuses to load - I had an engineer walk through the process with me, each of us doing the exact same steps on our respective computers, and it loaded on his but not mine).

Thank you so much for your help!

Koz means with -6 dB the recording level.
The normalization for export should be around -3 dB (Koz prefers generally -3.2 dB for Mp3 export).

If the recorded track has -6 dB and the RMS level is -20 dB, the outcome will be too hot. This is because the RMS level changes exactly the same as the peak level:
-6 dB peak → normalization to -3 dB
-20 dB RMS → is now -17 dB (not in range -18 to -23 dB).


The Audacity meters are measuring peak levels.

Setting levels for any recording is trade-off between setting the level low enough that you will never “clip” the signal, and setting it high enough that the noise of the recording process won’t make the signal worse. For voice having the meters bouncing up to -6 is a pretty good choice to be reasonable assured you won’t hit 0 (but it certainly possible if you get too excited at some point in the script.)

If you were recording a musical performance you would almost certainly keep some additional “head room” (-12 to -20 are common values). Because when the organist hits that C-major chord at the beginning of the Maestoso you don’t want your meters to go slamming into the red.

In post-processing the recommendation is to normalize the signal to -3 (or -3.2) dB. This is because ACX requires no hotter than -3. At this point there is no danger of clipping due to the dynamics of the performance. When you normalize to -3 the software is going to find the very highest peak in the entire performance and adjust the gain such that peak is at -3.

If, when you normalize to -6, you see the volume go down, the implication is that there were peaks somewhere in that recording greater than -6, which if you were recording with the meters bouncing near -6 is not surprising at all.

“RMS” represents power, if you ran a small electric heater off your signal I would be an indication of how many watts that heater would generate. As such it’s a somewhat better (but still quite flawed) indicator of “loudness” than the peak levels. ACX uses it as a standard to insure your loudness is within reason.

If you are having problems getting the plugin to load I’d recommend posting in the section of the forum appropriate for OS you are running (windows vs mac vs linux) and be sure to include the version of Audacity and of the plugin.

Koz prefers generally -3.2 dB for Mp3 export

It turns out that if you use the Normalize thing at -3.2, you are much less likely to go over -3.0 when you convert your work to MP3. MP3 is an imperfect format. It re-arranges the sound a little to get the smaller file sizes. Sometimes, for odd technical reasons, the sound volume actually goes up when you convert. This helps avoid offending the ACX Robot by going over -3.0 when you submit. The specification is not to go over -3.0.

-3.2 is not perfect, but it’s a good fuzzy rule to follow with almost no downside.

The MP3 conversion can still come out wrong and you should certainly run the works through ACX Check before you submit.


Yes, all the meters in Audacity are peak-reading except one. You can switch the meters from multi-colour to the legacy format with two bouncing green bars. The light-green bar to the left measures RMS or loudness, the darker green to the right peak. Right-click the meter.

It turns out the multi-colored meter is convenient (for me) to use during recording and the dual-green one for playback and production. I believe in 2.1.1, you can split the meters and have both.

Recording at -6 on the multi-coloured meter, once again, is a fuzzy goal. That’s where the colours shade to yellow and it’s easy to see out the corner of your eye what you’re doing.

If many of your show peaks around -6, then you are not likely to peak into the red area very much on theatrical emphasis (but you can) and you also avoid recording so quiet that you start having acoustic fist-fights with the system noise (fffffff). You must avoid peaking all the way up to 0. That is overload or clipping and creates harsh, permanent crunchy sound damage.