Traffic noise in location recordings...removal or minimization

[studer58]

Recording quiet sources..solo piano, sopranos in quartets, baroque trios, that sort of thing, in churches in the city area which have lovely acoustic spaces. Great...but what do you guys do about the inevitable traffic noise ? Not the constant background rumble, but more so the occasional bus or motorbike or ambulance/police siren or aircraft. Do you carve around it with high and low pass eq's, re-record the offended-against section again, use spectral surgical removal processing ? I have Wavelab 6 and Waves X Noise but until now haven't had to use them to save the day on some unrepeatable recording opportunities. I'm also using REAPER, and I'm sure that between these options there must be some which work better than others. Have any of you climbed and conquered this particular mountain ? Studio recording is not an option, it's a case of fixing already recorded material which has occasional patches of noticeable noise.
Thanks for any advice you may have on this.

[Modo]

Music tracks seem to be a little tricky to clean out noise you don't want. I've had some luck taking out mids and low mids on dialogue tracks. Sometimes only minimizing the undesired. You have already tried those waves clean up wizard plug ins? If that and eq can't get you where you want in your mix, then you need to record farther anyway from freeways or in a more secluded quiet place.

[Matti]

Spectral editing with Izotope RX and Algorihmix ReNOVAtor come to mind,
your Wavelab has something like that also but not as sophisticated

Matti

[Demokid]

This might be hard to do or it doesn’t work at all… Use a second microphone recording only the noise. Then phrase invert the noise track. It might cancel some noise out…

I’ve heard this is done if artists (I think Bono) don’t want use headphones when recording vocals in the studio. They phrase-invert the signal to the monitor used to play the music for the singer… Don’t know if it works.

I use the same technique when I buy new studio monitors which needs to be played for say 50 hours before they are “warmed up”. Be for I go to work in the morning I put the monitors in front of each other (front against front) and play white noise really loud. The only thing you hear is a quiet hiss due to phrase cancellation which doesn’t disturb the neighbours. That way I get over 8 hours of “warm up” every day. :D

[philper]

Originally Posted by Demokid

This might be hard to do or it doesn’t work at all… Use a second microphone recording only the noise. Then phrase invert the noise track. It might cancel some noise out…

I’ve heard this is done if artists (I think Bono) don’t want use headphones when recording vocals in the studio. They phrase-invert the signal to the monitor used to play the music for the singer… Don’t know if it works.

I use the same technique when I buy new studio monitors which needs to be played for say 50 hours before they are “warmed up”. Be for I go to work in the morning I put the monitors in front of each other (front against front) and play white noise really loud. The only thing you hear is a quiet hiss due to phrase cancellation which doesn’t disturb the neighbours. That way I get over 8 hours of “warm up” every day. :D

What you are describing doesn't work at all for normal recording--the acoustic situation in the hall is too complex. I will admit that many years ago I tried this technique (in desperation), it just made a bigger mess. But if you have the time....maybe you'll discover something. Otherwise, the best solution to that kind of disturbance is another take and some editing, with any sort of digital post-prod fiddling a distant second best.

phil p

[Gerax]

I have recorded in great sounding spaces with poor isolation, they were noisy, and I mean REALLY noisy, so I know where you are coming from. If it's not static or continuous noise (like AC, lamp buzz, distant traffic low rumble which can be dealt with pretty standard NR algorithms) your best solution is to record lots of material so that each section is covered at least 3 times, and then careful and patient editing. Random noises rarely happen twice in the same music spot. Pay particular attention to fading notes in codas and very low dynamic passages by turning up the headphone volume a bit more.
I almost NEVER have had perfect isolation so I'm pretty much used to it, but after editing you cannot hear a single noise.
Also, try to record at night whenever possible, so that there's less outside noise, though I understand it can be uncomfortable...

[WhiteSheets]

Originally Posted by Demokid

Then phrase invert the noise

I think you mean 'phase invert'

but that would only work if the noise during the required recording was the same as the recording of the actual noise - which would be unlikely to be the case, especially in the case of traffic, which is constantly changing in its frequency content.

[ysmgj]

izotopeRX has a "partials+noise" i think it is in the spectral repair section.

it can work wonders on small sections as it'll sample the bit before the noise and the bit after and duplicate it to make it blend. of course you need to play with the settings but spend a day on RX and you'll be sorted.

[7rojo7]

treat the take as the noise and work up from there (NR programs) otherwise record a lot of bkg noise and mix
get as close to the subject as possible
This and the notes you make during the take will be useful in ADR

[Matti]

Originally Posted by Demokid

This might be hard to do or it doesn’t work at all… Use a second microphone recording only the noise. Then phrase invert the noise track. It might cancel some noise out…
[FONT]

Not in the real world -and abouts this Bono theory, you got it wrong.
The story tells us they used the monitors in opposite phase, never tried here
but out of curiosity I tried to record voice with the rest coming from the control monitors and then record the rest without moving the mic or changing any levels ( and obviously mix with the other 180 )
It sort of works but not to an extend of being useful in real situations IMO

Matti

[7rojo7]

you can invert the noise if you have a clack, just slip the noise track to the dialog trk
bring up the noise faders when you want it to disappear, you could eveven control this with a gate
take at least double the amount of 'tone' to help the mixer mix

[7rojo7]

sorry
do the best you can with what you have
treating the music as the noise and making the smallest edits possible has worked for musical sources, if you're in PT, Audio Suite (Cedar is the best for this.
treat a larger section (a second before and a second after the culprit, then you can control the Xfades more precisely).
only your hairdresser will know
Algorihtmics for PC

[Matti]

Originally Posted by studer58

Not the constant background rumble, but more so the occasional bus or motorbike or ambulance/police siren or aircraft.

Well, been there and seen it done -even by me
Iz RX or ReNOVAtor both can kind of calculate something to the edits you´re removing, not to leave a hole in the spectrum.
I did a rock concert with frequent problems with feedback and after some fairly easy edits not noticeable if at all.
Done some classical concerts and all the coughs and like all gone.
The masters edit the performance with these tools, not me

Matti

[DannyL]

When working in cathedrals in the New York metropolitan area, we would record at night to reduce external noise. Also, use multiple takes.

And be sure to record at least 1 minute of room tone at least once an hour.
(All the mics on, everybody in place, no one playing or touching anything.)

If you need you can use this as a noise print. The more often you record room tone, the better the chance you have to make it work in your noise cancellation software. The noise print will definitely vary over time.

Recording too much room tone however will make everyone angry or bored.

Regards;
Danny

[studer58]

Thank you guys, it's comforting to know we all find ourselves in the same boat when less than ideal conditions prevail. It's useful to have a whole palette of repairs available, and wisdom comes from knowing which one is the best to use in which context (eg spectral editing vs inserting a few seconds from an alternate take, for example) I'll certainly check out Izotope and ReNovator too...
Seems to me a noise-proofed facility not too far from a major town or city could soon pay for itself in recording hires, as such places are increasingly hard to find in conventional churches, halls etc. It would need to have exceptional acoustics of course, and that means a lot of real estate lying around for the sole purpose of being "very quiet". Maybe 'dual purpose' is the way to go, with it being a restaurant, library, meditation retreat centre when it's not being a recording location !

[Matti]

When you have problems described PM me, I might be able to help you
withing reason

Matti

[Mark A. Jay]

I don't know if this sort of thing has been incorporated into PT, CuBase, et al, but this is a technique that has been used for years (usually to great success) in the signal-processing world such as in products from Bruel & Kajer, LMS, Head Acoustics, et al:

Kalman filtering.

Here's a brief review: Kalman filter - Wikipedia, the free encyclopedia

In practice, one needs to dedicate a phase reference (a transducer, i.e. microphone) to the 'noise', and this signal must be synchronously acquired with the other sigals (from the mics). Thus, in theory, you could dedicate a mic to the exterior of the space (likely requiring an armed guard) and use that signal as the phase reference.

This phase reference allows the Kalman filter to "focus" on that signal that shows up as bleed (from outside) in the other mics located in the venue, but because the algorithm has a phase reference of the offending noise, it filters out only that, leaving the rest of the time domain (recording) intact. In effect, it's a sort of 'smart filter' and knows only to remove (or attenuate) the signal that you don't want (the bleed from car exhaust for example) while leaving the signal that you do want intact.

As long as there is reasonable dynamic range between the levels of bleed and the levels of the performance (at the same frequency), then the technique works very well - it's used a lot in what is known as 'source decomposition' and allows you to extract harmonic content (and accelerating motor vehicles are replete with harmonic content).

In the signal processing world, this approach is sometimes used just to remove one source (while leaving all others intact), or in some cases, to allow the harmnic components that have been extracted to be 'massaged' in some way (i.e. level etc) and then put back into the original signal. This sort of thing is commonplace in the design of sound for things like automotive (exhaust / induction) as well as the consumer industry (washing machines, dishwashers, refrigerators etc).

[Matti]

"has been used for years (usually to great success)"

-How come all the manufacturers of NR soft / hard -ware try to solve
the dilemma differently to your suggestion.
I´ve been messing around sound bussiness some 35 years and yet to find a good universal solution other than close the channel,

Matti

[Mark A. Jay]

Originally Posted by Matti

"has been used for years (usually to great success)"

-How come all the manufacturers of NR soft / hard -ware try to solve
the dilemma differently to your suggestion.
I´ve been messing around sound bussiness some 35 years and yet to find a good universal solution other than close the channel,

Matti

...they are different 'environments', or rather, have been for years. Noise and Vibration / Sound Quality signal processing individuals don't usually work in the music industry, and vice versa. Consequently, there's often little awareness of a technique that was developed for one application area that might work really well as applied to a similar problem, but in a different setting or discipline.

The good news is that there now seems to be a lot of cross-pollination of disciplines; those well-versed in things such as DSP / convolution implementations (which are found in anything from cell phones, to radar, TV, and music) are adept and writing code for music production applications, or at least, have the fundamental knowledge of where to start.

The Kalman apporach is really software-only; as long as you have the signals synchronously acquired and have phase reference(s), then chances are favorable that the approach will work.

For signal processing (such as in automotive sound quality and product development), Kalman came out of necessity - like all inventions. In particular, it was designed to allow filtering when two or more sources happened to produce the same spectral content (frequencies) either at a particular moment in time, or for prolonged periods of time. If conventionally filtered, then both the desired signal and the undesired signal are both affected by the filter process; Kalman solves this issue. It's a big issue in the Noise and Vibration world, because often times, we wish to know 'how much of the offending signal must be removed (how much attenuation) before the problem is no longer perceived as a problem. This deals a lot with things like masking and thresholds...and this is how component noise is often assessed and modified.

The only time I have seen the technique stumble is when the magnitude of the tones (from different sources) are close to one another in level (a few dB apart). However, in the example we spoke of (bleed from engine noise), chances are that the dynamic range would be more than enough for the technique to work well.

[Matti]

Now give us before/after samples as your explanations goes over my language restricted head, so to say

Matti

[Mark A. Jay]

Originally Posted by Matti

Now give us before/after samples as your explanations goes over my language restricted head, so to say

Matti

...I once wrote a technical paper for this for a Congress of the Institute of Noise Control Engineering (INCE) entitled "Sound Quality Analysis Tools For Multi-Source Signals".

You can get the paper at: INCE-USA

The paper was presented at the Sound Quality Symposium, Dearborn, MI., USA, 2002. I'm not allowed (by terms of use) to post content from that paper on this forum, but you might be able to get a copy of it on-line.

I have not personally used this technique for recording purposes, - because none of my mixing / music production apps have Kalman functionality (that I know of or have been able to find) - I wrote this paper having used the signal processing tools that were available to me in my 'day job'; Kalman is relatively commonplace in signal processing software such as available from the vendors that I cited previously.

However, conceptually, the problem is the same one that I addressed in the paper, which is why I suggested it as a possibility that would likely work in your setting.

[boojum]

Originally Posted by Mark A. Jay

I don't know if this sort of thing has been incorporated into PT, CuBase, et al, but this is a technique that has been used for years (usually to great success) in the signal-processing world such as in products from Bruel & Kajer, LMS, Head Acoustics, et al:

Kalman filtering.

Here's a brief review: Kalman filter - Wikipedia, the free encyclopedia

In practice, one needs to dedicate a phase reference (a transducer, i.e. microphone) to the 'noise', and this signal must be synchronously acquired with the other sigals (from the mics). Thus, in theory, you could dedicate a mic to the exterior of the space (likely requiring an armed guard) and use that signal as the phase reference.

This phase reference allows the Kalman filter to "focus" on that signal that shows up as bleed (from outside) in the other mics located in the venue, but because the algorithm has a phase reference of the offending noise, it filters out only that, leaving the rest of the time domain (recording) intact. In effect, it's a sort of 'smart filter' and knows only to remove (or attenuate) the signal that you don't want (the bleed from car exhaust for example) while leaving the signal that you do want intact.

As long as there is reasonable dynamic range between the levels of bleed and the levels of the performance (at the same frequency), then the technique works very well - it's used a lot in what is known as 'source decomposition' and allows you to extract harmonic content (and accelerating motor vehicles are replete with harmonic content).

In the signal processing world, this approach is sometimes used just to remove one source (while leaving all others intact), or in some cases, to allow the harmnic components that have been extracted to be 'massaged' in some way (i.e. level etc) and then put back into the original signal. This sort of thing is commonplace in the design of sound for things like automotive (exhaust / induction) as well as the consumer industry (washing machines, dishwashers, refrigerators etc).

Brilliant and elegant. I will file that one away for future reference. Thanks.

[sonare]

Real-world is how Izotope and Algorithmix make a nice living saving our butts! I just completed clean-up on several masters and on one used both of those two plus Sequoia spectral cleaner. In one spot I used one "magic wand" to get rid of the artifact left by another one in banishing a really nasty sound. Sirens are in a class of their own-- just start another take.

Normal Izotope is simply amazing!

Rich

[JonesH]

Originally Posted by sonare

I just completed clean-up on several masters and on one used both of those two plus Sequoia spectral cleaner. In one spot I used one "magic wand" to get rid of the artifact left by another one in banishing a really nasty sound.

Heh, that's a nice one, Rich! Dual spectral cleaning

[sdog]

+1 for Izotope RX as a post, after-the-fact tool. I use it more for dialog than music, but you can really dig into the details of noise that you can't isolate with with just your ears or other tools. The "Spectral Repair" tool is indispensable.

Most of us can hear "a problem". But except for perhaps a few golden ears, we can't hear that there is a "10ms glitch at 362Hz". Because RX relies heavily on a visual component that matches what you hear, it is very easy to "see" your audio problem, isolate it, and usually repair it.

The repair itself can sometimes be a little noodly until you get a feel for how the control parameters affect the signal, but if you are patient (monstrous huge undo stack for plenty of trial and error) you can repair dang near anything.

[Matti]

Originally Posted by sonare

used both of those two plus Sequoia spectral cleaner

Rich

Remember that the Sequoia plug in is the Alghorithmix ReNovator to much extend

Matti

[Mark A. Jay]

Originally Posted by sdog

+1 for Izotope RX as a post, after-the-fact tool.

...The repair itself can sometimes be a little noodly until you get a feel for how the control parameters affect the signal, but if you are patient (monstrous huge undo stack for plenty of trial and error) you can repair dang near anything.

I just took a quick look. FFTs are indespensible, so it's good to see those are available (this is a fundamental tool in the signal processing world). I'm not sure how much control one has over the FFT parameters (line size, overlap %, window type etc) but from the looks of it, they have done some nice things, so there should be a pretty fair measure of parameter control. Sonograms are anotehr indespensible tool, so it's good to see those in there as well. For those who are familiar with the FFT process and its limitations, these parameters are intuitive and thus, like all apps, the most utility will be realized by those who understand Fourier analysis.

The de-noising (for mains hum and its harmonics) look like conventional notch filters, though they do in fact provide linear phase capability (which implies an FIR approach (a choice really) rather than an IIR approach).

I'm not sure they employ Kalman filtering though (which, as I mentioned in a previous post is extensively used in signal processing, so maybe these cats incorporated it)...I'll have to dig to see, but in any case, high-Q notch filters can solve a multitude of issues pertaining to mains.

What I see (on the web page) I like. Thanks for the tip. Any idea what the suite of tools costs?

[Matti]

Noise reduction and spectral editing are a graft -how good the programs used may be, try ( and adjust your thinking ) -take lots of your time before paying customers

Matti

[Alexey Lukin]

Originally Posted by Mark A. Jay

I'm not sure how much control one has over the FFT parameters (line size, overlap %, window type etc)

All this is adjustable in a Spectrogram Settings window: time overlap, zero padding, window size, window type, etc.

Originally Posted by Mark A. Jay

I'm not sure they employ Kalman filtering though (which, as I mentioned in a previous post is extensively used in signal processing, so maybe these cats incorporated it)

No, Hum Removal is just a set of fixed notch filters. However they have a "Learn" capability for detecting frequency of the hum.

[Mark A. Jay]

Originally Posted by Alexey Lukin

All this is adjustable in a Spectrogram Settings window: time overlap, zero padding, window size, window type, etc.


No, Hum Removal is just a set of fixed notch filters. However they have a "Learn" capability for detecting frequency of the hum.

All good in the hood then on the FFT stuff.

As far as hum removal is concerned, I take it they made it generic (for any sustained tone?) I mean, there's not much to "learn" in terms of hum from the mains - it'll either be 50 or 60 Hz and the respective harmonics depending upon where the recordings were made (i.e. 50 or 60 Hz mains). Conversely, any FFT with high enough spectral resolution would allow you to see this (narrowband 2-D FFT or a sonogram).

If, on the other hand the 'learning' involves finding a phase reference for the offending hum and its harmonics, then this would be akin to what Kalman filtering would do (scroll up for a description).

If it did that it would be extremely useful (for the reasons I cited), and dare I say, be 'the shizzy'?