Edit / speeding up production

[Eric Tomlinson]

If any can help, I'd appreciate it.

My recording desk is working well and my problem is now cleaning up the output.

Mouth noise is still a problem and I have been going through each file editting this out by hand.

By setting Audacity to waveform (db) I can easily spot them as I scroll through and they show up in the audio as a tiny spit noise.

I am willing, if not happy to do this, but I am going to feel a complete dummy if somebody on here says. 'Just set this filter to this and run it on the whole file.'

I am working with Audacity 2.3.3 on an i7 windows 10 laptop.

http://monkeyonmyshoulder.co.uk/wp/w...mouthnoise.wav


http://monkeyonmyshoulder.co.uk/wp/w...mouthnoise.png

I have diligently ignored any temptation to the EQ functions and in fact the last session required almost no tuning other than edits for spit noise and timing issues. (Big gaps / repeating the reading.)

I am still working forward rather than going back to rework the old files, as I am certain I will learn more as the project progresses.

Thanks for so much help so far in this.

ET

[Rick Reineke]

You could try a vinyl restoration tool that removes clicks and pops. iZotope's 'RX Advanced' would be my go-to. A less expensive RX Elements' version is available as well. In any case, set the sensitivity as high as possible without generating artifacts. Mouth noises that slip by the restoration processor can 'usually' be surgically removed using an audio "editing[" app such as Sound Forge Pro, which allows redrawing the waveform or other 'surgical' tools.

[Paul R Johnson]

Well - if it was me, I'd simply ignore it. I expected something horrible, or troublesome, or something people would spot as a major problem and I hear perfectly normal speech from a person who probably has a tiny gap in his teeth. I don't count it as a mic placement issue, or a fault, just how some people's voices really are. Listening to just the audio doesn't cause me to highlight it as a fault, and with picture, even less so. In fact, the only thing that went through my head was his accent, and I started to try to place it, forgetting to listen for the noise. I certainly would not be looking at trying to remove it with a tool, because it's part of his natural speech, so removing it would be a task and a half, and frankly, rather pointless. Plenty of pro voiceovers have far worse natural breath sound. If you listen to Sir David Attenborough or high profile VO artists, you hear far, far worse. Stop beating yourself up for nothing.

[Bruce Watson]

Originally Posted by Eric Tomlinson

...but I am going to feel a complete dummy if somebody on here says. 'Just set this filter to this and run it on the whole file.'

Don't know about spoken word, but this happens a fair amount with choirs and choruses. Some people make interesting noises when they talk/sing. It's what humans do. These clicks and pops usually occur way up around 9kHz or so. What editors often do in post is use a parametric EQ to make a narrow spike upward on the graph (to make the noise louder and so easier to hear and locate the correct frequency) that they can move back and forth to find the spot where the clicks/pops occur, then turn reverse the spike into a dip (to make it more difficult to hear) of some appropriate size (3-6 dB is usually sufficient).

The idea is to suppress it, not eliminate it. Trying to eliminate something like that usually leaves artifacts you can hear, if nothing else you can usually hear the "hole" you leave behind. So you just suppress it to the point that it no longer calls attention to itself.

So yeah, a small narrow dip up around 9kHz, applied to the whole file often does the trick. And often there are a couple of frequencies that need this. "Smacks" are different than "pops", and every speaker has them at somewhat different frequencies, etc. Just sayin' that YMMV.

[Andrew Smith]

That's a very good point about not eliminating sounds entirely.

In the parlance I am familiar with, one "attenuates" the undesirable sound rather than remove it entirely as the resulting audio 'hole' can leave it sounding not quite right. There is also the aspect of collateral damage by audio repair plugins to the tonality you wish to keep.

Andrew

[Rick Reineke]

My standard work flow when working with dialog is using high and low-pass filters, since there in little usable content in the human voice below 100Hz and above 10kHz. High frequency tics and such are generally in the the 8k and higher range.

[Patrick Tracy]

It might be tedious, but fixing each individual artifact is the best sounding solution. I generally don't apply an effect to the whole track to fix short duration imperfections. Sometimes a plugin like a de-esser or de-clicker can help, but often they'll start to affect parts of the track you don't want them to.

Most likely what you're doing is the best approach, but perhaps you need a better tool. My favorite for this is clip gain automation in Pro Tools, but Reaper has a pretty good version of that.

Here's what I did on your sample in Reaper (the image is the waveform just before "decisions"):

[Eric Tomlinson]

Thanks for the input guys. I have taken a week out due to slipped disc. I think the consensus is to keep doing it by hand. Every attempt to filter it does seem to mar other places.

[Greg Miller]

I've tried a few click removal plugins (on 78 & LP transfers). My experience is they need to be applied *very* carefully or they can leave low frequency artifacts.

As far as Eric's particular situation, I wonder if it would be possible to

first, split the frequency band so the offending frequencies are on one track, everything else on a second track;

next, apply some careful downward compression to the HF track, maybe around 10dB or so, and just slow enough that it would not raise gain on the mouth noises;

finally, recombine the HF and LF tracks.

[Patrick Tracy]

Originally Posted by Greg Miller

As far as Eric's particular situation, I wonder if it would be possible to

first, split the frequency band so the offending frequencies are on one track, everything else on a second track;

next, apply some careful downward compression to the HF track, maybe around 10dB or so, and just slow enough that it would not raise gain on the mouth noises;

finally, recombine the HF and LF tracks.

You could simply use a multiband compressor to do that without manually splitting and recombining the frequency bands. But I don't think it would work because the noises are probably not louder than other sounds in that frequency range. I bet you'd end up with what sounds like an overly strong de-esser, which makes people sound like they're lisping.

[Paul R Johnson]

YouTube (Short URL)

Sir David Attenborough - for many, the voice of the BBC, complete with breath noises, plosive sounds and some sibilance.

I wonder what he'd think about somebody editing out every little imperfection - his voice is quite famous, used everywhere and far from perfect.

Same with Stephen Fry and Harry Potter

YouTube (Short URL)

These have breathing noises and occasional sibilance. They could have removed these tiny imperfections, but they chose not to. Quality wise - I really hear little between the one talked about here and these big names.

[Greg Miller]

Originally Posted by Patrick Tracy

You could simply use a multiband compressor to do that without manually splitting and recombining the frequency bands. But I don't think it would work because the noises are probably not louder than other sounds in that frequency range. I bet you'd end up with what sounds like an overly strong de-esser, which makes people sound like they're lisping.

I'm sure the unwanted noises are much quieter. If I were trying this (which I do not have time for) I would set the threshold just a bit higher than the level of the mouth noises. Then I'd set the time constants so there was just a small delay before the gain went back up to unity, but maybe a somewhat longer delay before downward expansion happened. With luck you might get the settings so that it took care of most of the noises automatically.

Or perhaps it would be better to use frequencies other than the mouth noise frequencies to control the expansion. So if only mouth noise were present, the "mouth noise frequency band" level would be expanded downward. If any other frequencies were present (above some threshold, obviously) then gain would be returned up to unity.

I guess any of this would depend on how similar the various mouth noises were.

Now it's starting to sound like a challenge ... but I really do not have time right now!

[Andrew Smith]

I certainly understand the challenge of deep diving into solving a problem like this in the midst of other things that need attending to. Reckon this thread will still be here when you get a chance to have another look at the issue.

Andrew

[Eric Tomlinson]

Originally Posted by Rick Reineke

My standard work flow when working with dialog is using high and low-pass filters, since there in little usable content in the human voice below 100Hz and above 10kHz. High frequency tics and such are generally in the the 8k and higher range.

This has really cleaned up the sound. Thanks

[Eric Tomlinson]

End result so far. I am being less fussy about mouth noise, but rifle shooting the obvious lip smacks when it appears in a silence. Since using the clean up run filtering above from Rick, I am not adjusting any bass/ treble at all and noise reduction is almost redundant.


Greg ... I kind of follow what you are suggesting to split the tracks, but I think this is beyond my capability with Audacity as I stand today.

Please feel free to laugh, one full record session trashed due to starting too soon after eating and my stomach gurgling through 90% of it. No zips on clothes to catch on things, no curries. I am learning.

Appreciating and trying to absorb the suggestions.