?Breath removal/reduction recommendations?

[Mixedup]

Hi All.

This is for dialogue only.

I'm doing breath reduction with manual edits. I have got that process down to a single swipe-click in Reaper. That moves the breath to an adjacent track, which gives me loads of control in terms of setting levels, adjusting fades, truncating v long breaths, rolling off top end etc. V happy with the results, and it's pretty quick with practice — it's easy to spot breath waveforms and to do edits as I play through.

I'd save a hell of a lot of time and effort if I could find an app/plug-in that can attenuate breaths reliably, though! Thing is, I haven't yet found one I'm happy with. RX10, Accusonus ERA, Waves, Descript... they're all *nearly* there. About 85% maybe. But they all throw up enough errors, whether in terms of damage to wanted sounds or missing occasional breaths, that they don't really end up saving me time, and I have less co trol over the results. It's particularly the case when working with already compressed/bandwidth restricted sounds captured over VOIP or phone.

Not sure if this is a case of the workman needing more practice or the tools not quite meeting my standards of acceptability!

Does anyone else here have any experience of comparing such tools, and have any particular recommendations that I should check out?

All the best,

Matt

[The Elf]

I've not yet found an automatic process that works as well as manual, but multi-band compression (especially Waves C4/C6) is about as close as I've come to it.

My custom frequency split technique (discussed somewhere back in the mists of the SOS forums!) does a pretty good job, but still relies on a bit of automation, so not really much better than a multi-band really.

[sonics]

It sounds as if you're at the same stage as me. I've used all the tools you mention, but for some jobs manual editing is simply the most effective way.

It strikes me that this is a perfect candidate for AI processing. Is there not an online service that provides this yet?

[Tomás Mulcahy]

Have you tried the Izotope RX Breath Control module? It's included in the basic version. I've not tried that particular module, but RX is a must-have for audio post IMO.

[Mixedup]

Thanks guys. Yes, I have RX10 Pro. And Descript is a good AI based online tool but its Studio Sound feature doesn't offer separate access to debreath, denouse, dereverb etc. And it has a great transcribe and 'edit audio like a Word processor' function. (Not to mention autostripping filler words, repetition etc.). It's definitely where this world is headed. But as I day, there are always enough 'misfires' in all these things that I still have to go through, listen out for it and edit... so there may be fewer clicks and swipes, but really it takes just as long for not quite as good/controllable a result.

Wouldn't it be nice to be able to train an AI de-anythinger?!

I've tried multiband compression, dynamic EQ and a deesser too, Elf, including fiddling with the sidechain. But no beans with any threshold-based processors.

[James Perrett]

Wouldn't it be nice to be able to train an AI de-anythinger?!

The best ones are open source and trainable so we just need a tame computer science student to create an easy training method that the average studio person can use.

[James Perrett]

Wouldn't it be nice to be able to train an AI de-anythinger?!

The best ones are open source and trainable so we just need a tame computer science student to create an easy training method that the average studio person can use.

Actually it looks like the people behind Demucs are working on AI speech de-noising.

https://facebookresearch.github.io/denoiser/

They seem to be getting into all kinds of interesting work...

[Tim Gillett]

Except for maybe the second file in 'real rec' (0:17 secs) the sample speech files seem pretty much 100% intelligible, even before the noise has been reduced.

https://facebookresearch.github.io/denoiser/

[Mixedup]

Thanks. Yeah. As I say Descript's Studio Sound 8s AI based and does a really impressive job of it already. But success varies according to captured audio quality (over which you may not gave control/influence), whether any processing is baked in, accents etc.

But I'll look into the other suggestion. Ideally I want AI only for specific processes. Kinda like RX but with better results haha. Also, I'd like the moon on a stick please

[Hugh Robjohns]

Also, I'd like the moon on a stick please

Available from Amazon, apparently. Some assembly required....

[LeVzi]

Waves do a plugin called debreath.

It does work, but tuning it can be fiddly.

[MarkOne]

Also, I'd like the moon on a stick please

Available from Amazon, apparently. Some assembly required....

I did buy that, I wouldn’t recommend it though. The stick isn’t quite long enough, and I got into all kinds of trouble with NASA. Something to do with a near miss with Artemis 1.

[JAB Studio]

Izotope RX Dialog Isolate is the holy grail for dialog cleanup. I've tried everything and short of manual editing, this is absolutely the best.

jab

[Matt Houghton]

Thanks. Yeah, I have and use RX10 a lot. It's wonderful for lots of things. But I'm finding the breath control side of things is too hit and miss. Hence I'm doing it manually (or at least only semi-automatically!) in Reaper, and hence the original post.

[JAB Studio]

Not " breath control". It sucks. Use "dialog isolate". I master e-books and think dialog isolate is the bomb.

jab

[Matt Houghton]

Thanks. Yeah, I understand: Breath control is the aim, Dialogue Isolate the means you suggest to achieve it, instead of the dedicated de-breath tool. I've not had any great success with either for this, compared with manual edits. Maybe I need to out more time into learning how to get the best out of it...