Removing breaths from audio narrations

[maher]

I know this topic is sensitive and I know the audio narration world is kind of divided on whether to remove breath noise from audio recordings or not. From my modest research, it seems that the trend is going towards removing the breaths, or at least reducing the level of those loud breaths.

I am a digital signal processing engineer and I enjoy inventing algorithms that make people's lives much easier. I am in the final phase of developing one that can automatically remove breaths (or simply reduce their volume level) from audio recordings.

Current solutions like noise gates and debreath plugins need a lot of tweaking the various parameters to make them work. And after they"work", they still do not output good results (frequent chopping of letters, inaccurate inset and outset points, etc).

That is why most professional voiceover artists resort to manual removal of these breaths. Now, doing manual work for a one-minute commercial is no big deal. How about a 15-20 hour book? This is where the algorithm helps.

I am looking for a few voice artists who are willing to try the algorithm (which is still offline for now) and give me feedback.

Please be in touch if you are interested.

[Peter Why]

I usually manually remove or deamplify my breaths when I'm editing. I'll remove them if I think the timing is better without, and will deamplify if they're too obvious visually on the sonogram.

I'll be happy to test out your algorithm.

Peter

[maher]

Do you have an audiobook on the platform here? If yes, just point me to it. If you not, can simply upload a sample chapter on the forum. I will run the algorithm on it and send it back to you.

[Peter Why]

I uploaded a chapter a couple of days ago.

Section 14 in this book on Easter Island: viewtopic.php?t=96812

Peter

[maher]

I ran the algorithm. You have files at this GDrive Folder:
https://drive.google.com/drive/folders/1paic4TN9NYR5kALYjS_sDn7G-I2JdLqy?usp=share_link

One that eliminates more than 80% of breaths. And another that reduces their levels by 75%. In the first one you will notice a few letters (espcially s) from the end chopped off (5-6 in the entire 24-minute recording recording). In the 2nd one you will not notice that they are chopped off. Copying one a [chopped] letter from original track to another takes a few seconds using Audacity.

Manually removing (or reducing) the breaths on this 24-minute segment takes around 80-100 minutes. The algorithm takes about 48 seconds to return the clean file.

Let me know what you think.

Maher

P.S. I suggest you put the original on one track and the clean one on another in Audacity and then activate dB view. This will allow you to have a very good visual comparison.

[Peter Why]

It's late for me; I'll listen to them tomorrow.

I do have some completely unedited wav files, if you'd like to play with them.

Peter

[lightcrystal]

I know this topic is sensitive and I know the audio narration world is kind of divided on whether to remove breath noise from audio recordings or not. From my modest research, it seems that the trend is going towards removing the breaths, or at least reducing the level of those loud breaths.

I am a digital signal processing engineer and I enjoy inventing algorithms that make people's lives much easier. I am in the final phase of developing one that can automatically remove breaths (or simply reduce their volume level) from audio recordings.

Current solutions like noise gates and debreath plugins need a lot of tweaking the various parameters to make them work. And after they"work", they still do not output good results (frequent chopping of letters, inaccurate inset and outset points, etc).

That is why most professional voiceover artists resort to manual removal of these breaths. Now, doing manual work for a one-minute commercial is no big deal. How about a 15-20 hour book? This is where the algorithm helps.

I am looking for a few voice artists who are willing to try the algorithm (which is still offline for now) and give me feedback.

Please be in touch if you are interested.

I don't tend to remove any breaths. I have noticed that my plurals involve a breathy tail, for instance "shops". If I reduce the breaths then it becomes the singular such as shop.

I have done a Solo of about 6 hours: "Yorkshire Battles". You could take that and run the breath removal through it.

https://librivox.org/yorkshire-battles-by-edward-lamplough/

[maher]

It's late for me; I'll listen to them tomorrow.

I do have some completely unedited wav files, if you'd like to play with them.

Peter

Sure. If you send me the url for those wav files that would be appreciated.

[maher]

I don't tend to remove any breaths. I have noticed that my plurals involve a breathy tail, for instance "shops". If I reduce the breaths then it becomes the singular such as shop.

I have done a Solo of about 6 hours: "Yorkshire Battles". You could take that and run the breath removal through it.

https://librivox.org/yorkshire-battles-by-edward-lamplough/

The algorithm is smart enough in 90%+ of the time to leave those alone. BTW, you already do a good job at controlling your breaths. Neverthless, I ran the algorithm on section 4. Below is the urls to both files, the one that eliminates breaths (or rather 80%+ of them) and the one that reduce them.

https://drive.google.com/file/d/1dPSXAfED2EdKsbVjlMTeK4Qp0FMb73MN/view?usp=share_link

https://drive.google.com/file/d/1G5bZBqUERTvtsk0WApsJNqq1oNzrqR5E/view?usp=share_link

Let me know what you think.

[lightcrystal]

I'm not an audio engineer of any description. I'll probably break every testing fundamental. Anyway.

I took all 3 tracks of Yorkshire Battles Chapter 04, and placed them into Audacity: one raw, one that you eliminated all breaths, and one that you reduced the breaths. Then I had them in three tracks above each other. I listened to one only by muting the other 2. [Note I don't normally use Audacity but it's simpler to set this up than it is in Reaper that I normally use.]

I don't notice a lot of difference. I am not listening in headphones. I notice that in the reduced breaths track that my plural 'S' sounds at the end of words still remain intact. That is good.

If you want me to do anything further please let me know.

[maher]

I don't notice a lot of difference. I am not listening in headphones. I notice that in the reduced breaths track that my plural 'S' sounds at the end of words still remain intact. That is good.

If you want me to do anything further please let me know.

You did the right thing by placing them one on each track in Audacity. Yes, you do need a headphone to hear the difference. What you can also do is to activate the dB view on each track in Audacity (Audacity by default shows the linear scale). This way you can see the difference in a visual way.

[Peter Why]

I can't rely completely on my hearing to judge the effect, but certainly on using the dB display, the difference between partial and complete breath removal is visible, as some breaths can still be seen and heard. And the terminal S's that I checked seemed undamaged.

Congratulations on what seems like an effective tool. Would it be worth mentioning in the Audacity forums?

Peter

[maher]

Congratulations on what seems like an effective tool. Would it be worth mentioning in the Audacity forums?

Peter

Thanks for the suggestion.

[pumpkibee]

I, personally, like the breaths in the audio still. It feels personal, like someone is reading a story to me in person rather than something professional and detached. Also, in my case, I feel like my timing of my breaths tends to go hand in hand with my reading style. It just feels more like mom reading a bedtime story to me, or something along those lines if that makes sense? I like 'em, makes the recording feel more human lol

In my opinion it's different than voice acting in an animated project where you must remove breaths because it costs money to animate an inhaling scene during talking sequences, or sometimes even singing some songs (like rapping) where breaths would break the flow of the song.

Although this does sound like a very useful tool for those who do prefer to remove breaths! (I never knew it would be a sensitive subject? That's interesting to learn lol)

[GettingTooOld]

I think it is a fabulous tool which will come in very useful. I hope it's free and open and offline, but is all good. I spent last week writing a program, I tell it one video on U-tube, it makes a folder on the hard drive called by the U-tubers name, then populates it with very big thumbnails and information about videos and then makes a webpage on my hard drive I can browse and read descriptions, titles, look at pics, see views and select videos to download, without the infernal YOU-WILL-WATCH-ADS-AND-CANNOT-STOP-THEM-NO-MATTER-WHAT and so forth, they really drove me to it.

If your script can find the spots and mark them in time, perhaps they can somehow let people know where they are, for checking for S's and then get their input if they hear a miss, and then it can do the audacity thing, or rather, it would be easier for the program to skip it, as it already reads the file.

Do remember to make an out-take output file, with just breaths, for people who like making prank phonecalls