Noise reduction problem - how to NOT throw the baby out with the bath water...

[ElJayBronx]

I make and edit opera videos, and a company with which I work moved into a new venue this spring.

I own a Sony FDR-AX100 and the company rented two more so I can effectively do 4K three-camera video, then use Premiere Pro CC 2017 to edit a multi-camera shoot. It's very hard work but the results are much better than anyone expected!

The audio separation using on-camera mikes from the left and right (unmanned) cameras is much better than that I get from the sound engineer, and I would like to use it. The problem I have is that the theater has a loud air-handler system. Applying the newly released adaptive noise reduction feature (or, for that matter, the now-obsolete approach - I've tried both), much is improved - but in softer orchestral moments (CARMEN Act 3 Intermezzo with a duet for harp and flute as a good example) the instruments sound almost under water. All the examples of this feature and others are of much simpler noise problems (such as a narration with background noise, not singers on stage and an orchestra playing at variable levels).

If the only answer is to go through the video and apply an effect selectively I'm afraid they're getting the lovely audio separation and faithful recording WITH the air handler noise. If anyone has an idea, not too labor-intensive, that would help me with this project, I'm all ears.

[SteveG_AudioMasters_]

ElJayBronx  wrote

Applying the newly released adaptive noise reduction feature (or, for that matter, the now-obsolete approach - I've tried both), much is improved - but in softer orchestral moments (CARMEN Act 3 Intermezzo with a duet for harp and flute as a good example) the instruments sound almost under water.

There's no such thing as an 'obsolete' NR - and in fact it's the (far from obsolete) process effect that will work by far the best on a constant noise. The truth about this is a bit brutal, I'm afraid - if you get bubbly noises, then you've overdone it. The good news is that there's a better method of applying NR, and that's to use a multi-pass approach using different FFT numbers each time, and not taking too much out at a single pass - I'd say 4-5dB max, although you might get away with slightly more with higher FFT numbers. But no, you shouldn't need to do this selectively; in fact that's generally quite a bad idea with anything live, as it can be distracting. And get the cleanest recording of the air handler you can to use as a reference file - preferably without audience noise in it.

Sound recording live theatre productions is a pain in the butt. If it's something you're going to do in the same venue on a regular basis, I'd go for a completely separate micing system that's optimised for what you need - and be prepared to spend a bit on some higher-quality more directional mics. Almost invariably these need to be a little closer than locked-off camera positions, I've found, and you have to be prepared to experiment a bit. But optimising the sound at source makes everything else a lot easier to achieve.

[ryclark]

I entirely agree with Steve. Don't put the baby in the bath with the water in the first place.

You have obviously invested quite a lot of money to get great pictures. So it would be well worth your while spending a bit more (audio usually costs less than video ) to get the best sounding audio recordings that you can. It might even be worth while to take separate feeds of the sound desks mics and record them Multitrack to do your own sound mix after the performance. But it does take a bit more work, of course to get polished results.

[Bob_Howes]

As others have said, the best way to clean audio is to get a good recording in the first place.  If you're not happy with a feed from the mixer (and I can fully understand why having been down this road myself) if time, budget and permission allow, I'd try 3 Bartlett Boundary Recording Mics across the front of the stage.  https://www.bartlettaudio.com/products/boundary-recording-mic  Depending where you are, you may find you can rent them (or similar) from a theatre supply place.

However, your problem now is the noisy recordings you have in hand already.  As others have said, don't discount the old Noise Reduction (Process) effect.  It's capable of better results than the Adaptive route but takes some time and effort.

The underwater sound tells me that you've applied too much noise reduction in a single pass.  You're far better off to do 3 or 4 subtle passes, changing the FFT size between each.  (The FFT size is on the Advanced menu).

Find a patch of silence with just the noise, highlight this, then go to Noise Reduction (Process) effect.  Select 2048 under the FFT size and use the sliders to set Noise Reduction of 10-15% and the Amount of reduction at about 10dB.  Select All, then run the effect.

Go back to the Noise Reduction (Process) menu, change the FFT size to 4096, grab the noise sample, select all, run the process.

Keep doing this, upping the FFT size between each take until you've done the highest FFT (16,384).

If you want to try with fewer steps, delete the one or two lowest FFT sizes.

Anyhow, this should clean your audio a lot without bad artefacts on the sound.

[SteveG_AudioMasters_]

https://forums.adobe.com/people/Bob+Howes  wrote

As others have said, the best way to clean audio is to get a good recording in the first place.  If you're not happy with a feed from the mixer (and I can fully understand why having been down this road myself) if time, budget and permission allow, I'd try 3 Bartlett Boundary Recording Mics across the front of the stage.  https://www.bartlettaudio.com/products/boundary-recording-mic

This can work well*, but it can also give you some difficulties in terms of directional pickup; you are relying very much on phase delays, and unfortunately this only really works below 1kHz. With three PZMs (like the Bartlett mics) you have reduced the intensity differences considerably, simply because of the way they work, and their hemispherical pressure responses.

*So how do you fix it? Personally I'd only be inclined to use two mics, in a bit from the edge of the stage, and then I'd use a Blumlein Shuffler plugin to increase the apparent separation - read www.audiosignal.co.uk/Resources/Stereo_shuffling_A4.pdf. Fortunately there are ways to implement this easily, and they're free. If you search around, you'll find several free VSTs, but the best fix, by a long way, is to use the Open Ambience Project's 'SHEPPi' - Open Ambience Project "SHEPPi" Free Spatial Enhancer. There's only one downside to this, and this is that because of the way it's made, it can only at present be a 32-bit plugin, but it will work fine with Audition using jBridge ( for Windows ) | J's stuff. Since a lot of the singing action tends to take place centre-stage, it's probably worth having a shotgun pointing at it anyway, and perhaps you could even have Bob's third PZM and only fade it in when you need it - otherwise it's going to reduce the stereo you get significantly.

[ElJayBronx]

Wow! I'm heartened to have gotten so many great responses!

The company involved is a small New York City organization with a large heart and not a lot of money (I am also a sometime opera singer and appear with the company).

We have had several venues during our seven-year existence and have in fact landed in one that seems to be friendly and attract a good audience. However, there is no orchestra pit and as mentioned a significantly noisy air-handler system. They strain to pay the rental fee on the two extra cameras (but recognize the amazing results - I can work some magic with 4K images, panning and zooming on them in post-production, not just cutting back and forth between them). The audio engineer was extolled to me but then I found out that the feed he could offer was monophonic and sounds flat - and not in a good way. I do not feel qualified to meddle in his equipment and mic hookups, though...In past productions where we have had one camera only, the audio generally came off the mic on that camera; in the present setup I sit controlling that camera in the lighting booth, which is plagued by having a really loud air-handler register and the sound is useful for one purpose - to sync with the other two cameras. That's what gave me the idea of using the outer cameras for audio. You're right, it's not always ideal. There can be too much audience noise at times, similar to the problems with the video where suddenly someone close to the left or right camera positions gets up to leave the auditorium and inadvertently ruins what otherwise would have been a good camera shot.

That being said, I'm encouraged to try the multi-pass approach recommended here. It's a bit ironic that the older way is in fact more appropriate than the new "adaptive" methodology, but I'll work with it. And thanks to all.

I'm not sure putting PZMs on the stage lip would be a good idea. There's a lot of foot noise onstage and those mikes would be behind the orchestra, which is on a different level. I may mount better, more directional mikes on the outer cameras and see how that works. This go-around used the cameras' internal mikes since I had no notice I was using the signal from those cameras (assumed a better feed from the board; assumed that the higher-quality mic on the center camera would carry the day which it does not due to the booth air-handler register).

And thanks to all.

[mikerephone]

All good advice. I might add, if you have some money to spend - Adobe's built-in NR plugins are pretty good, but iZotope RX's are much better in my humble opinion. I've gotten amazing results using multi-pass techniques with that plugin.

~Dave

[SteveG_AudioMasters_]

mikerephone  wrote

All good advice. I might add, if you have some money to spend - Adobe's built-in NR plugins are pretty good, but iZotope RX's are much better in my humble opinion. I've gotten amazing results using multi-pass techniques with that plugin.

I agree, but I didn't think that it was quite politic to mention that here!

[ElJayBronx]

Just wanted to let you all know how well Steve's suggestion worked! What I had forgotten about the performance with the most noise: The system A/C failed that day and there were numerous fans aiming air into the auditorium which of course contributed to the noise. The multi-pass suggestion (applying concepts that I would have had NO other way to discover) worked admirably, reducing the noticeable noise not quite to zero but enough so the listener won't have the experience ruined. Thanks to all.

[SteveG_AudioMasters_]

Thanks for reporting back, and I'm glad it's worked out. Amazing what you can do with the 'obsolete' NR tool, isn't it?

[ElJayBronx]

I'm sorry about that, I'm just used to a lot of tools I'm used to in Premiere being supplanted and they're marked "obsolete". I note this is not the case with this one in Audition. I wish when they come up with something that's supposed to be "adaptive" (presumably better i.e. more intelligent) they should make clear its limitations. Clearly, it's not meant for my application (too complex, I guess). Anyway, the harp/flute duet is gorgeous and much less noisy. Thanks!

[SteveG_AudioMasters_]

Nothing ever gets marked as 'obsolete' in Audition - I don't think that the developers would dare. What happens is that every time something gets changed, altered or removed, there's a massive outcry from all the people that were using it that the developers didn't realise about. A good example here is the Playlist. At one stage it was removed, and we were all genuinely surprised at the numbers of people who came out of the woodwork and complained like stink about that! So back it went in the next version...