processing audio that includes long silences

I am making software training videos. Audio narration is recorded first (Audacity 2.3.0, Win10), then I play back the narration while doing a screen recording of me using the software. I know that certain operations performed by the software will take 15-30 seconds to complete, so while recording the narration, I leave an appropriate amount of dead air at the points where the software will be doing its thing.

I want to apply the audiobook mastering process to my audio recording, but only the parts where I am actually speaking. In order to do this, I split my audio tracks into sections of dead air (whenver it is longer than ~5 seconds or so) and sections of speech, move all the sections of speech onto an empty track, apply equaliztion, rms normalization, and limiting as per the process instructions, and finally move each section from the processed track back to its original location on its original track.

This works, but I am curious whether there might be a more elegant way to do it? Some recordings might have 8-10 one to two minute tracks with 20 or more segments, so there can be a lot of splitting and moving pieces around.

Or if there is not another way, just confirmation of that would be very reassuring and helpful.


You’re right that the first and last steps in mastering don’t care in the least how long the presentation is, but RMS Normalize works on a fuzzy average and long stretches of silence may throw the results off.

I think I follow that. You do that so you can apply Mastering to one single track. How are you selecting the individual snippets to put them all back?

I can think of one thing that might help. We’re holding a Batch or Macro would would allow you to apply one single process and do all three Mastering jobs. It’s a bit brittle and things can go wrong, but it may be valuable for you.

I need to wait until I get back home.


Hi Koz,

I actually do equalization before cutting everything into pieces because it also makes the dead air sections sound better. I do RMS Normalize on the bottom track in the picture, and since it is already there, I also run the Limiter on the track. I use the Time Sift tool to move the snippets from the bottom track back to the original positions on the tracks above. Kind of like doing a jigsaw puzzle.

These are my first recordings since your ACX process was published, and it works like magic. I would be interested in trying out the macro if you think it would help. I understand if it does not yet have all the kinks ironed out.


Here is the Macro (Batch) file written by Deborah Nolan, but it’s the earlier version without the signature.

MasteringMacro.txt (191 Bytes)
Download it and slide into the Macros folder in your Audacity System Files. I have zero idea where that is on your machine. It’s the same folder that has the Plugins folder that you needed to add filters and effects to Audacity.


Open or create a simple sound performance.

Effect > Equalization

Which preset do you have? Low rolloff for speech? This is one of the program defects. If you don’t have Low rolloff for speech, Select Curve and pick Low rolloff for speech. You should be in EQ Type : Draw. Click on the curve somewhere between 2000 and 3000 and drag the smallest possible change up or down.

Screen Shot 2019-04-18 at 11.51.21.png
Note the horizontal is two thin lines now instead of one.

You can’t hear that change, but it will cause Low rolloff for speech to stick through multiple applications.


Back to your test sound. Select the whole thing by clicking just right of the up arrow.

Apply the three mastering tools like you have been doing and watch what happens to the blue waves when you do. Inspect with Analyze > ACX-Check. Edit > UNDO back to the original performance.

This time Tools > Apply Macro > MasteringMacro. OK the error message.

You should get exactly the same change you had doing the tools the long way and the sound should pass ACX Check.

So from now on you can use the Macro instead of all the individual tools.


I was going to ask you what would happen if you applied the one correction to all the individual tracks, but you can’t do that because you need to select all the tracks and selection fails if you use Time Shift Tool. At least it does on mine.

Try it. Hold the shift key and try to click more than one track.

I could force it to work by making multiple, very special, custom tracks, but that was almost as much work as what you’re doing.


I have been using the Low Rolloff for Speech preset.

I followed the instructions for changing the EQ curve between 2000 and 3000 Hz. If I understand correctly, the purpose of this is to make the default unnamed curve essentially the same as the Low Rolloff for Speech preset (and doing this is related to the error message that you mentioned), and it is needed so the macro will work, but would not be needed in general when doing the procedure manually one step at a time.

I installed the macro - the macros folder on Windows is C:\Users<your username>\AppData\Roaming\audacity.

Performing the steps manually and using the macro returned identical values in the ACX Check results panel, so thank you. It will be handy to have the macro available.

Selecting more than one track and applying RMS Normalize causes the tracks to change so that all tracks are the same length - it adds sections of zero signal (don’t know the correct terminology) to the previously empty portions of the tracks.

Both RMS Normalize and Limiter take twice as long, probably because it is processing all of both tracks as opposed to different sections of each track.
ACX check returns two results, both of which say the noise floor is ~-300 dB, likely because it is looking at the zero signal sections.

It is not a problem for me to move everything onto one track in order to process it. Thank you for the macro.


“Mastering” is something that you do after all of the editing is complete, and the show has been mixed down to a single track.

the show has been mixed down to a single track.

True, but this isn’t normal. The presentation, as I understand it, is small sections of dialog amid longer than normal sections of silence. The sort of thing that might throw off RMS Normalize, designed for conversational flow.