Best Practices for removing breath sounds in a natural sounding way from voice overs?

[Dolmetscher007]

I am an Instructional Designer, so I create loads of eLearning courses. That means I do a lot of voice-over recordings of myself explaining how to do something to do with some piece of my company's software.

I am the new guy on the team, and they hired me specifically because of my experience with recording/mixing audio. However, 95% of my audio experience is with recording music. I've done some audio for film before, back in college. But I am far from an expert in voice work for web-based eLearning courses.

My team is made up of six people, non of which have the slightest idea of what they are doing when it comes to capturing good audio. They are all very smart and capable people who know the software product amazingly well, but they all just use USB Yeti mics, sat on their desk, in some room within their house, just talking into the microphone. I'm sure the grain and polarity knobs are in the same place they where when they were given the microphones to take home. So, my manager is asking me to see what I can do to improve the overall sound quality of everyone's recordings. I don't want to open that whole entire can of worms in this thread; I'd like to just start with breath sounds.

I do all my editing in Logic Pro X. This is my workflow:

• I navigate to their .mp4 video file in the Browser section of Logic, and double-click it to load the video and extract the audio to track one.
• First thing, I use the Normalize tool to normalize the whole track.
• Add a Channel EQ instance that I basically use to chop of all really high and really low frequencies.
• Add the Expander plugin to ease out any background noise like laptop fans, or air-conditioning/heaters running in the background.
• Add a Noise Gate and try to fiddle around with the settings to make it so that the gate is closed when they take a breath between words, click their mouse, or smack lips... but that it opens and closes naturally around the words I do want to hear.
• I then add a Limiter on the main stereo output so I can really squash the louder sounds down without having to do any track automation to keep things smooth.

Overall, it all makes a drastic improvement in clarity and in making each video from person to person sound more homogenous. But... the noise gate is really an issue. When it closes... there is a noticeable void of all sound. If you're wearing headphones, it is slightly tiresome after a while to hear the sound switch so abruptly from on, to completely off. To get around it, I have a track of just... "room tone" that I recorded myself, that I let play in the background at a very low level, so that when the gate closes, you hear at least the slight sound of the dead room with mic recording that I loop every 30 sec.

I guess my biggest issue is... I don't really know of any place where I can go and hear what "good voice over tone" really is. With music it's simple. I know exactly which engineers are known for what kind of sound, and what band, song, or album to turn to as a reference to have something to shoot toward. But I have no clue who makes corporate videos with excellent voice over work. Mine may be fantastic, but then again, I may be shit at it.

I realize this thread is a little sprawling. My apologies. It's all a little stressful.

[Sam Spoons]

I don't use Logic but when I use a gate I don't want it to shut off all unwanted sounds but to reduce their level. The gates on my X32 have a Gain Reduction parameter which I set to reduce the noise but not to remove it completely.

We have a couple of people who do a lot of this kind of stuff, I'm sure they will be along presently with better/more detailed advice.

Welcome to the forum

[Luke W]

It sounds like you're along the right sort of lines already. If the other people on the team are going to continue to record their own voices then it may be worth looking into the possibility of giving them all some pointers in how to get the best results. Obviously with everyone working in different and probably less than ideal places it'll still need some work to bring it all together at your end, but it could certainly make your life easier!

Overall, it all makes a drastic improvement in clarity and in making each video from person to person sound more homogenous. But... the noise gate is really an issue. When it closes... there is a noticeable void of all sound. If you're wearing headphones, it is slightly tiresome after a while to hear the sound switch so abruptly from on, to completely off. To get around it, I have a track of just... "room tone" that I recorded myself, that I let play in the background at a very low level, so that when the gate closes, you hear at least the slight sound of the dead room with mic recording that I loop every 30 sec.

I'm not a fan of gates for this sort of work personally. I find that they're too brutal a tool to do the job properly, and the time you spend getting them to deliver acceptable results is usually better spent working manually. As you've noticed, silent gaps sound very unnatural so the best approach is usually to reduce the level of breath/mouth noise rather than get rid of it completely. We're really used to hearing people speak so you can get away with far less than you maybe could with a lot of other sounds, hearing someone deliver an entire paragraph without breathing sounds rather strange!

My approach when working on voiceover work is to go through manually, and manually reduce the levels of any noises that I feel are being emphasised too much. I usually work in PT and tend to split the region I want to work on, sort the levels with Clip Gain and then use crossfades to blend it back with the voice. I don't use Logic very much so there are probably some tricks I don't know about, but I'm sure the approach is very similar.

If there's any pops/bangs/noises that can't stay, I do very much like you've described with your room tone track. I usually pinch it straight from the track I'm working on so it matches without too much messing around, and then paste it in as needed.

Hope that helps.

[Hugh Robjohns]

Breathing is a natural part of human speech, and removing it completely resist in a sound quality that is both very natural, and gives the impression of bad editing!

There's nothing wrong with applying some downward expansion to make breath sounds less obvious, but if the mic is placed sensibly and with good gain structure breath sounds really shouldnt be a major problem at all.

So, rather than spending hours and hours finessing and applying complicated signal processing to every single voice over for ever more in an attempt to fix poor source recordings, I would suggest investing some time instead to helping your contributors optimise their own local recording setups so that you have better source material to work with which will require much less 'fixing'.

That would typically mean getting you VO artists to move back a little from the mic, and placing the mic to the side, rather than directly in front. A windscreen might also be helpful.

Being further from the mic will reduce the signal/noise ratio, letting in more ambient sound, so some attention could be applied to reducing external unwanted sound by using makeshift broadband absorbers around the speaking position, such as hanging duvets etc.

[Drew Stephenson]

2 suggestions,
1) check if your gate has a reduction option, try just taking 30-40db out rather than reducing it to silence.
2) next time Izotope RX Elements is on sale for $29 pick up a copy for the voice de-noiser. It's an excellent tool for speeding this kind of stuff up.

[Aled Hughes]

next time Izotope RX Elements is on sale for $29 pick up a copy for the voice de-noiser. It's an excellent tool for speeding this kind of stuff up.

And that time is right now!

I was transferring an old 78 record today. Had a look too see if the job could cover the cost RX Elements for me... and I got it for £17!

Merely loading the de-clicker on the track was a vast improvement on my previous efforts..

[James Perrett]

I'd echo some of Hugh and Drew's suggestions...

A good mic technique is vital and should be the first thing you try to correct. The stand on the Blue Yeti may well be too short so it would be worth seeing if you can raise the mic up to mouth level. As Hugh says, place the mic off to one side a few inches away from the mouth.

I would do the bare minimum of standard processing. A high pass filter is probably sensible with a turnover of around 80Hz and, if Logic allows, you could normalise to a standard LUFS level (I notice various podcasts seem to use -24LUFS) but that's probably all that I'd do. I certainly wouldn't normalise peaks to 0dBFS. Everything after that would be dependent on the particular voice.

Drew's suggestion of Izotope's Voice Denoise is good and is often available as part of the RX Elements package at $29 (plus VAT and duties). Don't be tempted to go cheap here - all the other alternatives won't be in the same league as RX - it is extremely good. Multiple passes of mild noise reduction are better than one pass of drastic noise reduction. With RX I use 10-12dB at most for each pass and rarely need more than two passes.

Where more than one voice is involved it might be necessary to eq the voices so that they don't sound glaringly different. It my also be necessary to use compression but I'd choose the compression settings to suit the voice rather than use a single setting for everything. You'll probably find that you end up with channel settings that you can save for each person which can then be easily recalled in the future.

[CS70]

I am the new guy on the team, and they hired me specifically because of my experience with recording/mixing audio. However, 95% of my audio experience is with recording music. I've done some audio for film before, back in college. But I am far from an expert in voice work for web-based eLearning courses.

My team is made up of six people..

I would be surprised if there is a significant difference. You (them) need to control their recording space. Perhaps the difference is that voiceovers use effects much more subtly, if any, and there is nothing to hide in, so the control must be even tighter. For the same reason, the sound of the mic, mic technique and the vocal timbre of the speaker are even more important. James Earl Jones isn't Darth Vader's voice because of his looks.

In addition, many companies tend to select people from how much "mouth noise" they make (there's usually a short, recorded audition), exactly because the more they do, the more editing is needed and turnaround time are often very important in that business. (I have some friends who did that kind of job and told me of their experiences).

If the people and the equipment are set, what you can do is to teach them to control their recording space and perhaps rudiments of mic technique. Something like a duvet behind them (I made a post on this to avoid rewriting every time), possibly hidden behind a background if video's involved, would do wonders, as much as telling them to put a scotch tape marker where they put their mic and their elbows so that they have a reference to maintain consistent distance. Tell them to have the microfone away from monitors. If you feel like splashing, buy them reflection filters.

the noise gate is really an issue. When it closes... there is a noticeable void of all sound. If you're wearing headphones, it is slightly tiresome after a while to hear the sound switch so abruptly from on, to completely off. To get around it, I have a track of just... "room tone" that I recorded myself, that I let play in the background at a very low level, so that when the gate closes, you hear at least the slight sound of the dead room with mic recording that I loop every 30 sec.

There's a couple tricks here. You normally don't want the noise gate to use a 100% range (or whatever it's called on yours). The attack, hold and release are also very important. Of course you want a good gate, with all the parameters that allow you to control it surgically. FabFilter G is pretty good, for example.

For very short noises, a good option is to forget the gate, cut the noise out and use Elastique or a good "stretching" tool to tug a little the clips left and right (obviously you want to cut "silence only" clips, short of stretching the entire dialog before and after) to cover up the hole. If you crossfade, the result is undetectable.

The room tone is a good idea for larger holes, but even better is to cut a section of good "silence" from another part of the clip and use that. Always crossfade.

I guess my biggest issue is... I don't really know of any place where I can go and hear what "good voice over tone" really is. With music it's simple. I know exactly which engineers are known for what kind of sound, and what band, song, or album to turn to as a reference to have something to shoot toward. But I have no clue who makes corporate videos with excellent voice over work. Mine may be fantastic, but then again, I may be shit at it.

Well, simply go and listen.. TV commercial ,radios, YouTube. Pick up the tones you like. With voice - especially with voice - you will know when it's good, but there's gazillion examples all around you, much more than music actually!

EDIT - just saw James' post and it's excellent advice.

[Tim Gillett]

A sound recording is worth many words. Any chance of a sound file example of the problems you mention? Making it sound good live (maybe not perfect) is usually far better than applying all sorts of processing later. In my audio book production days we couldn't do any post processing. Any and all processing had to be live. It was a good discipline.

[James Perrett]

I guess my biggest issue is... I don't really know of any place where I can go and hear what "good voice over tone" really is. With music it's simple. I know exactly which engineers are known for what kind of sound, and what band, song, or album to turn to as a reference to have something to shoot toward. But I have no clue who makes corporate videos with excellent voice over work. Mine may be fantastic, but then again, I may be shit at it.

This paragraph didn't register with me until I saw CS70 quote it...

Here are a couple of my attempts at recording decent voice overs. The artist is extremely good with a very controlled delivery and I did virtually no processing - just editing. I didn't create the final product but the voice sounds little different to the files I sent.

https://www.youtube.com/watch?v=GOCDaSpJNOM

https://vimeo.com/333738616

Of course there are plenty of other examples from the BBC and other broadcasters.

[Mike Stranks]

I am doing this sort of stuff 4 or 5 days a week... often poor and lossy recordings made on a smartphone in a poor environment. Some of my regular readers are quite 'breath-y'; others don't seem to have the problem.

Mic technique definitely does help as does correct mic placement. But I know too well the challenges of working with those who aren't either proficient or that bothered about getting good quality recordings.

IZotope RX has been mentioned. If you/your employers are serious about this stuff I'd recommend investing in the 'Standard' product. Not cheap, but for not outrageous money, 'simply the best'.

I will deploy some or all of the following if necessary:

Voice Denoise
De-Mouth Click
De-Ess
Reduce Breaths
De-reverb - carefully!
Nectar 3 Elements which can auto clean-up a voice. But I've found like so many plug-ins that less is more.

For breaths and clicks it is quite common for me to go though the recording reducing breath volume by 50% and editing out mouth clicks if I have time and the piece is short. If I don't have time and the piece is long and littered with noise then RX is the best automated processing software I've encountered - by a very long way.

You might want to explore 'Cleanfeed'. It's free and enables you to establish a real-time high-quality link to a contributor. You can this get them to read a few sentences and you can detect problems before recording. The recording is made by you, the call initiator. It does mean you have to sit there while they record, but if you' and your organisation is serious then it's a very good way of getting good quality and discovering issues before anything is committed.

There's more, but that's enough to be getting on with...

[Mike Stranks]

Just a quick 'add-on'....

I'll often Normalize a recording that's very low-level at the start of the process just so I have a better waveform to work with and I can hear what muck and mush is sitting there along with the voice. BUT I never Normalize to 100%. I' always leave myself at least 3dB of headroom.

You might want to explore 'Broadcast Processor': https://vst4free.com/plugin/548/

... but it's Windows-only...

I use it often. On it's gentle settings it's a good way of managing levels in a track (or mix) in an automated way without introducing horrible artefacts.

[The Elf]

... but it's Windows-only...

Also 32-bit only.

[CS70]

... but it's Windows-only...

Also 32-bit only.

That may not be a problem depending on the DAW the OP uses. I use 32 bit plugs all the time

[The Elf]

... but it's Windows-only...

Also 32-bit only.

That may not be a problem depending on the DAW the OP uses. I use 32 bit plugs all the time

Of course, but it needs pointing out. Many of us left 32-bit behind long ago.

[Mike Stranks]

I've started a new threadhttps://www.soundonsound.com/forum/viewtopic.php?f=16&t=76327 on the 32-bit/64-bit aspects so we don't wander too far away from the topic here !

[Tim Gillett]

Breathing is a natural part of human speech, and removing it completely results in a sound quality that is both very unnatural, and gives the impression of bad editing!

I agree, and breath is often a way of expressing and communicating various emotions and reactions. Editing speech down to just the words spoken seems like a risky business to me.

This show reel of audiobook narrations is interesting. In the sample narrations some breath gets through but there seem essentially silences between the spoken words, while by contrast the voice introducing each narration has natural breath preserved.

https://youtu.be/dsiZhXVRvJc

My impression is that in audiobook production there seems a tendency to smash flat any hint of background noise, even when it can be part of the narrators' performance.

[MarkyC]

Thanks for this thread guys. I wont add to the audio comments as people who know more have already spoken. Just some observations.

With things being what they are the use of videos with a decent vocal is such a valuable tool in business. My work has changed quite a bit recently because I have a decent microphone, access to a duvet and vaguely understand mic technique (through trial and error).

Whilst trying to train others is a great concept, I found that in my organisation, people just don’t get the importance of a decent vocal until they hear it played back against something of quality. Many think a decent sound just magically happens and blame their tools rather than making small changes that really help.

Most people say “I hate the sound of my voice” when what they really mean is “I hate the way my voice sounds when recorded badly”. They will spend ages putting a software demo together and then fail miserably when it comes to the audio.

Unfortunately what that has meant is that I have become the go to voice over guy (actually something I enjoy so no big deal). The workflow we have developed is that they do the video recording and talk over it. I play that back with Word Dictate on. I then take that Word Doc and make it a script. Import their file into Cubase, split out the audio, redub their part, export audio. Then take the original video and new audio file and import that into Camtasia, add any corporate titles (same intro to everything makes it more cohesive) and render.

It does take time but the payback is worth it.

By the way thanks for the heads up on, just purchased iZotope it.

[merlyn]

A constant and unrelenting stream of words is a popular style with the kids, as exemplified by e.g. Philip DeFranco, whose videos are kind of like the Nine O'Clock News for the internet generation :

https://www.youtube.com/watch?v=PNW_ZT1oVBY

If he took a breath the yoofs might click on something else.

[CS70]

Whilst trying to train others is a great concept, I found that in my organisation, people just don’t get the importance of a decent vocal until they hear it played back against something of quality.

Then I would just do that.

Engineering types are very sensitive to anything regarding their chosen skills, but also tend to disregard completely stuff they think it's simple or not worth of their attention. So make it about their ability to produce a complete quality product. You say these people are smart. It's the stupid ones which are harder to convince, so you should have a good starting point.

[MarkyC]

Yep - I think you are rite they do see it as simple and get quite perplexed when they get the better polished result so they just give up.

I'll try again on some of the easier to influence one