naturerecordists
[Top] [All Lists]

Re: Noise removal technique described (was: Filtering out jet planes

Subject: Re: Noise removal technique described (was: Filtering out jet planes
From: "Rob Danielson" danielson_audio
Date: Mon Apr 13, 2009 10:01 am ((PDT))
At 9:15 PM +1000 4/12/09, Andrew Skeoch wrote:
>Hi Rob, Vicki and all,
>
>>Andrew might care
>>to comment on whether his DINR filtering would retain beak snaps
>>below 500 Hz? Certainly Andrew is an absolute master at filtering,
>>much better than I am!
>
>>I might be wrong, but I believe Andrew is rolling off only under a
>>100-125Hz and using DINR in a manner more similar to parametric
>>filtering than brick-wall filtering. Rob D.
>
>I probably haven't described my technique very well, sorry Rob, no
>it's not parametric or brickwall filtering. I hope this clarifies...
>
>Vicki's description for highpass filtering and ambience replacement
>is a suitable one where there is no frequency overlap between noise
>and subject Unfortunately, much vehicle or aircraft noise will
>overlap with low-freq birdsongs, fundamental harmonics, mammal calls,
>woodpecker drummings, bill-snapping, wingbeats...
>
>So any effective processing will have to differentiate between noise
>and subject sounds. Brickwall or parametric equalisation will achieve
>limited results, as filtering will act on both noise and subject.
>This is where DINR is so versatile. DINR is a noise reduction plug-
>in, but it has more control and versatility than most other nr plugs
>available. What I'm describing here may only be applicable to this
>software, I'm not sure how it could be done using other plugs, sorry.
>
>The way I approach cleaning up say, aircraft noise, is to:
>
>1. Assess the frequency that the noise extends up to, or at least is
>lost among dominant subject sounds. 1000Hz is my usual default
>frequency.
>
>2. Create a multitrack editing session, and a master raw audio file.
>I'll often run a 100Hz brickwall filter on this file, just to remove
>the lowest rumbles.
>
>3. Create a processed version of the raw audio that leaves sounds
>above 1kHz unaltered, but reduces ALL sounds below 1kHz to the
>amplitude of the ambient background (determined from a noise-free
>section of the recording).
>
>This is a drastic and unconventional thing to do, essentially
>attenuating all sounds below 1kHz to the ambient level, leaving only
>a smooth noise floor below that frequency. This is the task only DINR
>can do, and I can't easily explain the settings required to anyone
>unfamiliar with the plug-in (It involves manually setting a noise
>floor curve, then using the 'Hi Shelf' function, setting the shelf
>freq to the lowest, 20Hz, thus specifying all the higher freq
>spectrum, and set it to attenuate, rather than the usual boost. I
>usually run a couple of passes on the same curve, attenuating by
>10-15dB each time, until an even noise floor is all that remains.
>I've included a screen-shot of the DINR settings for this with the
>accompanying files)
>
>4. The next stage is restoring the removed data below 1kHz, choosing
>between wanted sounds (birdcalls, bill snaps, wingbeats and so on)
>and unwanted noise (the louder components of aircraft, vehicle noises
>etc).
>
>Going back to the un-processed source recording, I set a noise floor
>threshold in DINR that captures only sounds that are below 1kHz, and
>ABOVE the actual noise (aircraft or vehicle noise) level. Attenuation
>is now set to infinity. Thus the only sounds let through this active
>filter are the louder sub-1kHz signals. By careful manipulation of
>the threshold parameters, the above-threshold sounds can be trimmed
>from the below-threshold noise floor. In practice, this may need to
>be done almost sound-by sound, as a series of small edits, and can be
>quite time-consuming. What you get is a series of disconnected tones,
>grunts, bloops and plops. At the end you may choose to bounce all
>these edits down to a single track.
>
>5. Mix the two processed files together 1:1, so adding the sub-1kHz
>components into the +1kHz/noise floor track.
>
>The result is full-frequency natural sounds, set against a clean
>noise floor throughout the frequency range, without any hint or
>residue of low-freq man-made noises. Any sounds below 1kHz are there
>because you can hear they should be, and are as complete as the
>source recording allows. As I said, it is a painstaking process,
>especially with soundscapes that may be tens of minutes long.
>
>I have prepared a set of files to show how this works in practise.
>I've put them on our website, as they seemed to be too big for the
>group's files section - download them from http://
>www.listeningearth.com.au/DINR/
>
>The recording is of Grey Francolins in Gugarat, India, with a jet
>aircraft going directly overhead. It is not the dirtiest recording,
>but it illustrates what can be done to improve things. My processing
>frequency for this recording was below 1200Hz.
>
>Files include:
>1. The raw audio file (with 100Hz brickwall filter),
>2. A simple attempt at filtering, using a highpass at around 500Hz,
>with two parametrics attenuating major jet noise bands at 250Hz & 350Hz.
>3. The audio processed using DINR as described.
>
>I've also included sonograms of each file so you can see what is
>happening. The interesting thing to note is a dove calling around
>800Hz, well inside the noise band. You can see how effectively it has
>been retained in the DINR processing, whilst being compromised by the
>eq filtering. Lastly two screen shots of the DINR settings for steps
>3 & 4.
>
>I hope this makes sense and is useful.
>
>Cheers,
>
>Andrew
>
-
Hi Andrew--
Thanks for the useful explanation and samples. Your technique has
done a great job of creating a longer take of the Grey Francolins
with no interruption that I detect.

When I adjust the adjust the playback level for a comfortable
playback level of the closer birds, the jet fading-in is not
tremendously objectionable to me. Albeit in a less ideal manner, the
jet does enhance my sense of the acoustic space, the recording
location. There might even be more spatial cues with the lowest
octaves in tact. I tried to locate the spot via satellite imagery and
the landforms look intriguing.

To me, what is most interesting is that we are making responses to
"noise."  Underlying this response there could be a personal balance
between taste (finding what sounds/feels good) and documentation
(finding the best version of "what" that happened).

In recent years, I've found myself been moving my mics further and
further away from loud sound sources in order to allow more sense of
the sound horizon along with nuance of the loud communications. Of
course, this makes the "noise" in the recording even more evident and
long takes unavoidable. I feel the wider tonal range tends to give
recordings a little more depth and make them a little "easier on the
ears" when played back at levels necessary to discern details.

Thanks again for following-up with samples that bring clarity to the
discussion. Rob D.

--







<Prev in Thread] Current Thread [Next in Thread>
Admin

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the naturerecordists mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU