Topic: Audience Noise Removal - Audacity 2.2.1

[rockphantom]

Howdy,

I've successfully captured the Whitesnake/Dead Daisies show at the Moody Theater in Austin, Texas last night.

I've listened to my recording for a few minutes this morning and was pleased with the results.
However, there was an obnoxious screaming woman behind me who was going off like a cat in heat for the duration of the performance.
I'm a relatively new Audacity user and was wondering if there was a way to remove audience noise (banshee like screams) without damaging the source material.

If you take a look at these two Audacity screen shots you can easily see the peaks produced by these screams.
Note: I merged and normalized the WAV files to -1.0db.

If someone could please provide me with the information I need or point me in the right direction, I'd appreciate it.
I'd like to remove as much of this noise as possible before sharing the recording.

Thank You!

[EarlyMorningRain]

sorry I do not use Audacity but perhaps the approach would be the same ??

Using Sound Forge, I zoom in on the suspect area(s), highlight the offending material (the lady behind you) and attenuate down a couple of db's or so. Then give it a listen and tweak and needed. Then wash/rinse/repeat the process for any other affected areas.

But certainly wait for others to chime in with their first hand knowledge before taking my advice.

[dean]

sorry I do not use Audacity but perhaps the approach would be the same ??

Using Sound Forge, I zoom in on the suspect area(s), highlight the offending material (the lady behind you) and attenuate down a couple of db's or so. Then give it a listen and tweak and needed. Then wash/rinse/repeat the process for any other affected areas.

But certainly wait for others to chime in with their first hand knowledge before taking my advice.

This is my approach using Audacity.  It's clunky, at best, but works.  I'm hoping another Audacity user has a better process...

[Life In Rewind]

sorry I do not use Audacity but perhaps the approach would be the same ??

Using Sound Forge, I zoom in on the suspect area(s), highlight the offending material (the lady behind you) and attenuate down a couple of db's or so. Then give it a listen and tweak and needed. Then wash/rinse/repeat the process for any other affected areas.

But certainly wait for others to chime in with their first hand knowledge before taking my advice.

This is my approach using Audacity.  It's clunky, at best, but works.  I'm hoping another Audacity user has a better process...

Agree with above posts...pretty much the only way of handling it.

You'll want to do this BEFORE you normalize...by looks of it - she's the loudest thing...

[dean]

sorry I do not use Audacity but perhaps the approach would be the same ??

Using Sound Forge, I zoom in on the suspect area(s), highlight the offending material (the lady behind you) and attenuate down a couple of db's or so. Then give it a listen and tweak and needed. Then wash/rinse/repeat the process for any other affected areas.

But certainly wait for others to chime in with their first hand knowledge before taking my advice.

This is my approach using Audacity.  It's clunky, at best, but works.  I'm hoping another Audacity user has a better process...

Agree with above posts...pretty much the only way of handling it.

You'll want to do this BEFORE you normalize...

Without question, do this before normalizing.

[rockphantom]

Thank you all for your replies! A friend suggested I do hard limiting of the master WAV prior to normalizing. Should I attempt to eliminate the screams first?
I've never attempted to hard limit a recording.

[Gutbucket]

Quickly reducing the level during the offensive sound is about as good as you can do easily.  If the the visible peaks far louder than the music are your caterwauler, you can easily reduce those to the same level as the music and make them at least less invasive, probably without introducing other obvious audible artifact.   You may be able to reduce a bit further, but the reduction will become audible as the music level suddenly decreases, sounding much like a limiter cutting in (which is what it is- a manual peak limiter applied just to those instances).

With more work you can sometimes do some interesting tricks-
If mostly on one side, you can copy the opposite channel and do a quick cross-fade to two-channel mono and back.  That can be less invasive than reducing signal level during the wail, since only the stereo imaging is effected and not the level of the music.

Similarly you can use Mid/Side techniques to cross-fade to more Mid-dominated mono or Side-dominated mono if the offense is in the center or the more diffuse ambience.  There is a Michael Gerzon paper from 1990 about various kinds of cross-fading tricks to do this, back in the analog mixing era, by setting up routings so a single fader could be rapidly moved in real-time to more easily reduce pops, dropouts, and these kind of unwanted noises during a dub with minimal artifacts. http://www.audiosignal.co.uk/Resources/Outside_the_mix_A4.pdf

The best modern approach to this are probably noise reduction commonly tools referred to as spectral editing tools.  They allow one to draw envelopes around offending noises on a spectral display that plots frequency against time, and apply clever algorithms to partly reconstruct the cut-out frequency bits from the unaffected surrounding time/frequency material.  Izotope RX is probably the most well known, can run standalone or as a VST I think, and is rather pricey.  Samplitude has this capability but is an full featured audio editing program.  Sony offers one I think, and I'm sure there are others, including very expensive tools from CEDAR and the like.  These are fantastic tools for this kind of thing which can near magically remove coughs and other noises, but they are complex and tend to be pricy unless included in the editor's built-in tools.

[thunderbolt]

I use hard limiting when stealthing at sit down festivals.  Play around with the settings.  I can get it to a point where I can pretty much highlight the whole set and squelch the cheers and claps to a very tolerable level.  This is for jazz concerts, however, which is a different thing than loud rock.

[Gutbucket]

Thank you all for your replies! A friend suggested I do hard limiting of the master WAV prior to normalizing. Should I attempt to eliminate the screams first?
I've never attempted to hard limit a recording.

 
If the loud screams represent the highest signal portions of the recording, you are manually limiting them when you zoom in and reduce their level.  A properly set limiter does the same thing, but may also effect the sound of other portions.  Setting it properly can often times be more difficult than just zooming in and drawing volume envelopes, or selection the region and reducing level, or however that's accomplished in Audacity.

Normalize last, after the noises are fixed to your satisfaction and everything else you want to do is done, other than tracking.

[morst]

I would say leave it alone.

The best thing you can do is to forget what she looks like, and come back to the master in a year with some perspective.

[voltronic]

Be very careful when using the hard limiter in Audacity.  I used to use it to knock down clapping, but for anything else I'd advise against it, as it is "haircutting" (clipping) the waveform, making your screamer softer but now horribly distorted. 
(Not to pull this OT, but here's how I knock down applause, and it sounds much better than a hard limiter.)

While maybe not the best either, you could try using the compressor.  Set the threshold to just above the level of the surrounding music, and then un-check the "make-up gain" setting.  This will also compress the music underneath the screamer, but if it's brief it might be an OK.  You could combine this with MS processing as Gutbucket suggests if the screamer is in the middle of the image, and you might do less damage to the music - squash the middle; leave the sides alone.  That's beyond Audacity's capabilities though...

Gutbucket is right that Izotope RX is the tool that allows you to really fix these things - the Spectral Editing he's describing is very powerful.  It's not all that cheap, but I find RX to be well worth it for what it can do.

If you'd be willing to post a clip, I'm sure a few of us here would be willing to have a crack at it.

[thunderbolt]

I should clarify.  The beauty of jazz is that most of the time there is a short gap before applause.  I hard limit what is essentially clapping/screaming, not music.

[voltronic]

I should clarify.  The beauty of jazz is that most of the time there is a short gap before applause.  I hard limit what is essentially clapping/screaming, not music.

I have abandoned hard limiting even for clapping, because the clipping distortion is so audible (at least if you use Audacity's hard limiter).  I only attenuate clapping using one of the methods I describe in that thread I linked from a couple years ago.

[aaronji]

I should clarify.  The beauty of jazz is that most of the time there is a short gap before applause.  I hard limit what is essentially clapping/screaming, not music.

There is that after solo applause sometimes, though, while music is still going on...

I use the "Hard Limiting (process)" in Audition to deal with this problem too.  Fast attack and release times.  Works very well unless you really over-do it.  I prefer this to envelopes because the levels of the underlying music are unchanged.

[rockphantom]

Thank you everyone for your advice.

Here is a FLAC sample of the second song of the night.
Screams at the beginning and end.

https://www.sendspace.com/filegroup/kO%2BzvX0CMs4psvg3xV1KlA

I can post a sample of a tune where the woman is screaming every 30 seconds if anyone is interested.

Here is lineage of my recording: Soundman OKM II Rock Binaural Omnidirectional Microphone - > A3 Adapter preamp -> Sony PCM-M10 (Line In) -> 24/96 WAV

[voltronic]

Thank you everyone for your advice.

Here is a FLAC sample of the second song of the night.
Screams at the beginning and end.

https://www.sendspace.com/filegroup/kO%2BzvX0CMs4psvg3xV1KlA

I can post a sample of a tune where the woman is screaming every 30 seconds if anyone is interested.

Here is lineage of my recording: Soundman OKM II Rock Binaural Omnidirectional Microphone - > A3 Adapter preamp -> Sony PCM-M10 (Line In) -> 24/96 WAV

I gave it a shot - the thing that worked best in Audacity for me was to use the Compressor with a threshold of -24, leave the other settings default, and disable make-up gain.  The screaming banshee is at least then no louder than the surrounding music.

I don't know if this recording is worth saving though, but that's no fault of yours.  The PA is so horribly over-driven - bass and kick drum are clipping it all over the place.  I've been to too many concerts like this recently - it really grinds my gears when FOH engineers can't do their job correctly.  You can make it loud within reason without ruining things.  Grr.

[rockphantom]

Thank you for the information voltronic.

Which Audacity settings to you recommend for the normalization of the recording?

[rockphantom]

One of my closest trading contacts who has been editing audience recordings for years, played with the sample and recommended that I try these settings to emphasize the music over the banshee screams.

1) left channel = hard limiting -17db
2) right channel = hard imitng -13db
3) normalizing = -0.05db

I should mention that he uses Abode Audition exclusively.
I can't find an equivalent for Audition's Hard Limiter in Audacity's Effect list.
I'd appreciate it if someone would give me direction.

Thank You!

[voltronic]

I played around with your file a bit more.  Try this:

1. Select the whole thing, and open the Hard Limiter module.
2. Set your dB limit to -24, Wet level 1.0, Residue level 0.2.  (Note - the dB limit takes a bit of trial and error to make sure your limiting what you want but not harming the music.  Residue level controls how "hard" (low number) or "soft" (high number) the limiter is.  See the quote from the manual below for more info.)
3. Hit OK to accept the changes.
4. Again make sure everything is selected, then go to Effect > Normalize.  First two boxes are checked; third is not.  Set your max level.  I leave this at -0.2 dB though if you're going to do additional effects or processing some would prefer lower.

Banshee lady should now be about equal in level to the highest peaks in the music.  To reduce her any more, you'd probably need to split to two mono tracks and try further limiting on the left channel, as that appears where she's hanging out most of the time.

From the Audacity manual pages:

==dB limit==
The level above which the input signal is processed. If the Wet level and Residue level settings are left at default, the peak level of the result will be clipped down to this level.

==Wet level==
The amount of the clipped signal that is fed to the output. It thus acts as a volume control for audio below the dB limit. If left at default level of 1.0, all sounds below the dB limit are fed to the output. At settings below 1.0, the volume of the result will be reduced.

==Residue level==
This allows a proportion of the signal that has been removed by clipping to be added back to the output. By default the value is 0.0 and so all of the clipped signal is discarded. By raising this level, some of the clipped signal will be restored, making the limit softer. More peaks and troughs of the original waveform will be retained, and the resultant volume level will be higher than that set in the dB limit.
{{Note|'''Special effect:''' If you turn the Residue level full up to 1.0 and the Wet level full down to 0.0, you will hear only the pieces of waveform that are above the dB limit you choose.}}

[rockphantom]

Thank you again, Voltronic.

Unfortunately, my installation of Audacity 2.1.1 does not include the hard_limiter_1413.dll plugin.
I've searched and can't find it online.

From what I've read, it's been replaced by a new effect called "Limiter".

http://forum.audacityteam.org/viewtopic.php?f=46&t=86465

Could someone please share the hard_limiter_1413.dll plugin with me?

Once I obtain the original Hard Limiter plugin I'll apply the  settings recommended by Voltronic to the entire recording and report my findings.

Thank You!

[voltronic]

Thank you again, Voltronic.

Unfortunately, my installation of Audacity 2.1.1 does not include the hard_limiter_1413.dll plugin.
I've searched and can't find it online.

From what I've read, it's been replaced by a new effect called "Limiter".

http://forum.audacityteam.org/viewtopic.php?f=46&t=86465

Could someone please share the hard_limiter_1413.dll plugin with me?

Once I obtain the original Hard Limiter plugin I'll apply the  settings recommended by Voltronic to the entire recording and report my findings.

Thank You!

I just checked and I'm running version 2.0.2.  I didn't realize they had removed / replaced it.  The new Limiter may work better for what you're doing, but I've never used it as I don't have that version.

Anyway, the DLL from my installation is here.  Put it in your Plugins folder and you should be set to go.

[rockphantom]

Thank you voltronic.
Unfortunately, the Justbeamit link for the hard limiter doesn't point to the file.
A typo perhaps?

[hoserama]

I generally hate doing any dynamics based solutions for crowd noise, as you're apt to get a lot of side effects with the rest of the recording.

I played with the beginning of your clip in Izotope RX3 and did spectral repair work. Also used the declicker with dialed in settings to take out those immediate claps.

I find the spectral repair is a much better solution...but is considerably more labor intensive.

Give it a listen:
https://www.sendspace.com/file/w3b1te

[rockphantom]

I generally hate doing any dynamics based solutions for crowd noise, as you're apt to get a lot of side effects with the rest of the recording.

I played with the beginning of your clip in Izotope RX3 and did spectral repair work. Also used the declicker with dialed in settings to take out those immediate claps.

I find the spectral repair is a much better solution...but is considerably more labor intensive.

Give it a listen:
https://www.sendspace.com/file/w3b1te

Wow!!

This is a night and day difference!!
I wish I had Izotope RX3 and the skill to replicate your work over the course of the show.

I need to find a Whitesnake fanatic out there with the skills to edit out the banshee screams...

Thank you hoserama for showing me what can be done!

[voltronic]

Thank you voltronic.
Unfortunately, the Justbeamit link for the hard limiter doesn't point to the file.
A typo perhaps?

Sorry about that.  I tried a different file sharing service - and updated the link in my post - this one should work.

[rockphantom]

Thank you voltronic.
Unfortunately, the Justbeamit link for the hard limiter doesn't point to the file.
A typo perhaps?

Sorry about that.  I tried a different file sharing service - and updated the link in my post - this one should work.

Thank you voltronic, I was able to download and install the hard limiter plugin.

[2manyrocks]

Of course it's too late for this recording, but if you find yourself next to a screamer and have the option to move, it could be a good idea to move elsewhere. 

[hoserama]

I generally hate doing any dynamics based solutions for crowd noise, as you're apt to get a lot of side effects with the rest of the recording.

I played with the beginning of your clip in Izotope RX3 and did spectral repair work. Also used the declicker with dialed in settings to take out those immediate claps.

I find the spectral repair is a much better solution...but is considerably more labor intensive.

Give it a listen:
https://www.sendspace.com/file/w3b1te

Wow!!

This is a night and day difference!!
I wish I had Izotope RX3 and the skill to replicate your work over the course of the show.

I need to find a Whitesnake fanatic out there with the skills to edit out the banshee screams...

Thank you hoserama for showing me what can be done!

I can 'lend' you a copy of RX3 if you want and give you quick crash course on it. That said, you've gotta be willing to put in the hours of work. Anyways, it's a much better pathway IMO than dynamics manipulation.

[rockphantom]

Here is a sample of the recording using the hard limiter settings that voltronic recommended.
Same track as the original sample.



https://www.sendspace.com/filegroup/j7jvveA0g4qp08FzZuWfrA

[aaronji]

Instead of applying the limiter to the whole file, I often find that it works better to zoom in a bit and limit small sections to the "local" maximum level of the music.  For some noises (such as a close clapper), I will often limit individual peaks.  Of course, this can be extremely tedious, so it depends on the nature and frequency of the noises in question...

[Gutbucket]

From the Audacity manual pages:

==Residue level==
This allows a proportion of the signal that has been removed by clipping to be added back to the output. By default the value is 0.0 and so all of the clipped signal is discarded. By raising this level, some of the clipped signal will be restored, making the limit softer. More peaks and troughs of the original waveform will be retained, and the resultant volume level will be higher than that set in the dB limit.
{{Note|'''Special effect:''' If you turn the Residue level full up to 1.0 and the Wet level full down to 0.0, you will hear only the pieces of waveform that are above the dB limit you choose.}}

A note on the note I've bolded above:  Doing that may be very useful in checking your threshold level setting (AKA "dB limit" in the terminology of this plugin).  When set that way you will only hear the portions of audio which are being limited and not anything else- just the high level excursions above the threshold setting.  Try setting the wet and residue controls that way and then play around with the dB limit setting until you hear mostly banshee yodels and applause spikes without much else musical leaking through, then revert wet and residue to their correct settings. 

Even if you prefer setting the threshold by ear in the typical way, I recommend doing this just to get a clear idea of what the plugin is doing by being able to actually hear what it is "throwing away".

[rockphantom]

Instead of applying the limiter to the whole file, I often find that it works better to zoom in a bit and limit small sections to the "local" maximum level of the music.  For some noises (such as a close clapper), I will often limit individual peaks.  Of course, this can be extremely tedious, so it depends on the nature and frequency of the noises in question...

Thank you for the suggestion.
 I will attempt limiting individual peaks during an upcoming weekend when it's too hot to do anything outdoors.

[mfrench]

I would say leave it alone.

The best thing you can do is to forget what she looks like, and come back to the master in a year with some perspective.

This is so true in so many cases.  Time passes and things sound differently.
Typically, I run slight compression on applause to bring it down to the level of the music; but, in my case, I'm dealing with an audience that is more mature than Rock, and, they don't make noises during the music. So, encapsulating the applause is a simple matter.

[Gutbucket]

Yes, time away from it and returning later with a fresh ear, unsoiled by recent bad associations is always a good idea.

..they don't make noises during the music.

Ha! I don't believe it!!

The banshee cry of the plastic cough-drop wrapper, the incessant program rustler, the woman with all the jangly jewelry that keeps messing with her hair, the purse rummager, the hard-breathing wheezer-geezer, the loud and out-of-rhythm foot tapper, the loud whisperers, the premature always-early and over-eager clapper, the end-of-each-movement clappers (maybe even every long rest), the outright snoring old man..  I know them all to well.  They are ubiquitous, attending every "quiet audience" event, everywhere, everytime.  If fortunate, they aren't seated immediately next to you or your mics, but they are there somewhere in the audience, guaranteed. 

[voltronic]

Yes, time away from it and returning later with a fresh ear, unsoiled by recent bad associations is always a good idea.

..they don't make noises during the music.

Ha! I don't believe it!!

The banshee cry of the plastic cough-drop wrapper, the incessant program rustler, the woman with all the jangly jewelry that keeps messing with her hair, the purse rummager, the hard-breathing wheezer-geezer, the loud and out-of-rhythm foot tapper, the loud whisperers, the premature always-early and over-eager clapper, the end-of-each-movement clappers (maybe even every long rest), the outright snoring old man..  I know them all to well.  They are ubiquitous, attending every "quiet audience" event, everywhere, everytime.  If fortunate, they aren't seated immediately next to you or your mics, but they are there somewhere in the audience, guaranteed. 

You forgot the classical concert cell phone offenders.  I was at a concert with the St. Olaf Choir (one of the world's best, look them up if you're not familiar).  In the middle of a very complicated Bach motet, someone's phone rang about 10 rows in front of me.  And rang.  And rang.  Either the person was deaf or didn't want to admit it was theirs.  And of course this was in a large church with beautiful acoustics.  Finally, the conductor stopped the choir, slowly turned to the audience, and said very quietly "Is everyone's phone off now?"  And he restarted the piece from the beginning.

Another time I was at the Philly Orchestra, and a woman's phone rang, and she picked it up and started talking on it.  This was also when you could buy those aftermarket batteries with the rainbow LEDs that blinked in circles while you were on a call.  [facepalm]

For clapping, I plan to print copies or post this for all of our school concerts this year.  Check out the website if you appreciate classical music humor, though most of it is piano-specific.
http://euge.ca/2012/12/30/applause/

[Gutbucket]

Ha! thanks.

[ghibliss]

You could remove the offensive sounds using a spectral editing tool which is in Izotope RX and a few other editing packages.  You can locate the frequencies you are trying to remove looking at the spectrogram and carefully edit it out.