Removing breaths

I’ve narrated an ACX book and have to merge in prerecorded commentary from the author. Can I lean on your expertise (again!) for some help? I have two problems:

    • His voice has a huge dynamic range, and “normal” compression gives an RMS of -26dB or so. I can use [X] “Compress based on peaks” to handle that, along with threshold/floor of -32/-60, but…
  • He is also very breathy, and compression based on peaks makes it even worse.
  • Fortunately it has a low noise floor, so I have room to play.

I think I have 3 choices here:

    • Nothing
  • Buy Waves Debreath VST plugin at $150 and hope it works
  • Hand-edit the breaths out (we’re talking approx 1 hr of inserts.)

I’ve attached a 10-sec sample that illustrates both problems. Breaths at 1.5 and 6.0, peak at 3.7 secs. Results:
As is: RMS = -34.1
Compressed: -22.6, Normalize => 25.8
Compressed based on peaks: -20.5, Normalize => -23.7

Are there any effects or plugins to help that I’ve overlooked?

I think your best bet is going to be to use one of the limiters (eg “broadcast limiter III”) to do a soft clip on the peaks in place like "toNALity ". With that the compressor does a pretty good job of bringing the rest into line with moderate settings. I don’t see anything objectionable in the speaker’s breaths even after healthy amounts of compression are used.

If you decide to use a limiter, I’d suggest that you try this one:
limiter.ny (4.32 KB)
This limiter will be included in the next release of Audacity. It was designed for Audacity 2.1.1 (or later), but I think that it should work correctly in 2.1.0.
Instructions are in the “alpha manual” (the manual for the next version of Audacity):
I’d suggest that you try the “Hard Limit” setting (note that this does not “clip” the audio - it is a true “limiter”.

Instructions for installing: Missing features - Audacity Support

(The “broadcast limiter” (all versions) is a “soft clipping” effect, and is similar to the “soft clipping” option in the attached plug-in).

Thanks for the limiter suggestion. Between that and your opinion that the breaths are acceptable, I can make this work. For the benefit of anyone else reading this, here is my processing chain:

  1. Amplify (to give the limiter something to bite on)
  2. EQ (LF roll-off for speech; this cuts my noise floor considerably)
  3. Limiter as suggested by flynwill
  4. Compressor: -20, -50, 3:1, 0.2, 1.0, Y X
  5. Normalize to -3.2

Thanks for that. I’m using 2.1.1 rc2, so it was there waiting for me, but the link to the docs assured me that the settings I picked were good.

Thanks again for the insanely great support on this forum :smiley:

For my breaths I use noisegate plug in. And I tried it on the sample with the gate threshold set to -30 and it worked fine. It works for me with my settings for recording volume on all but the most breathy breaths, but they’re easy to see and de-amp individually. I do it as I edit out the red spikes that I put in when I need to repeat a word or phrase.

But as a previous poster has noted, the breaths here are not so distracting (maybe they are somewhere else).

I’ll try the limiter, too and see if it’s an improvement on the noisegate that I’m using.

and here’s the result with what I believe are the normal settings for attack and release (100 and 500).

For my breaths I use noisegate plug in.

The performance is claimed to have wide theatric dynamic range and noise gates tend to fall apart without a level playing field.

That’s one of the “easy to dig yourself a hole” tools.


PhilMayes -

I have had some success with removing breath sounds using the following technique:

  1. Highlight a breath sound - i usually choose the biggest one I can find easily
  2. Click on Effect-Noise Removal
  3. In the new window that pops-up under Step 1 there is a button that says “Get Noise Profile” - Click on it. This identifies what you are looking at making changes to.
  4. Now that you have a sample of the sound (breath) you want to remove, select the section of audio (I choose the whole recording) you want to remove all breath sounds from and go back to Effect-Noise Removal and when the window pops up, under the section Step 2, click on the OK button.

I may have to do this a couple times to get the majority of those sounds removed and may even have to individually remove the breath sounds by muting/generating silence but those are far fewer after I have done this.

I am working on my 6th audiobook and this seems to work for me. I am going to try a couple of the options some of the people mentioned while replying to your post and compare the difference.

I hope this helps.

Best of luck.

Obviously you can do that, but the Noise Reduction trick only works on breath sounds exactly like the one you captured in the Profile step. That’s why it seems a little scatter-shot in application, or worse, it could only suppress part of a breath. It could also give you “holes” in your dialog where a breath is natural and expected…but it’s missing.

It’s not apparent in the posting, but if you have a partially successful filter tool, you need to read down through the whole show to make sure it worked. This gives you the “Times Five” rule. It takes you five times the length of the show (or more) to actually produce the show. I wonder how many people would think twice about being there for two and a half hours to finish up post on a half-hour show. That number is remarkably accurate.

I’m curious if you decided to suppress your breathing sounds because somebody complained about it, or you just thought it was a good idea?

I know of nobody who has been rejected by ACX for breath sounds, but I know plenty of people rejected for “Excessive Processing.”


Gee, what a lot of discussion in my absence :slight_smile:

I’ve just finished adding an hour total of author to 13 chapters. I ended up doing it the old-fashioned way - select the breath, Amplify -9dB, repeat (Ctrl-R) for all breaths. Do it twice on noisy breaths.

Adding this into the recording was a wild experience. The author’s voice is thin, with lots of energy in 2KHz - 4KHz, and a huge dynamic range. This pitch is around max sensitivity of the human ear. To make the two voices sound at a comparable level, I had to use something like: me: -20.5dB RMS, author -22.5dB RMS. Cross fingers and hope ACX approves!

It took mucho processing as well. I used flynwill’s suggestion of Broadcast Limiter III.

You just have to hit RMS higher than -23dB (and lower than -18dB)…after making the MP3. Doing all this at the WAV step will only get you in the ballpark. That’s the other reason skimming along on the edge of conformance is a shaky idea.