Tricks for peeling white noise out of audio?
Doc
music. Looking to strip out as much of the "shhhhhh" of the water as
possible but leaving the music as intact as possible. I've been using
an app that works with noise samples and am getting a certain degree
of success but wonder if there's any tricks to refine the results
further. I'm experimented with varying degrees of eq at various freqs
but find that a given freq seems to reduce one aspect of the hiss
while leaving others intact, and also is likely to peel out some of
the music as well.
I seem to be finding that using the shortest possible sample of noise
in this case seems to yield better results than using longer samples.
Thanks for all assistance.
Gr...@ao1.com
You can roll off the high end.
If that's not enough damage, you can peak it just below the roll-off.
If your listeners are really oblivious, use hiss-reduction or noise
reduction. No more hiss.
Soundhaspriority
"Doc" <docsa...@yahoo.com> wrote in message
news:23271079-ec0c-490d...@21g2000yqj.googlegroups.com...
Doc, I'm afraid that "a certain degree of success" is all you can get. There
are two basic classes of noise reduction: low pass filtering, and noise
shaping, which tries to reduce the noise without altering the power spectrum
of the signal. But it's a curious fact of noise shaping that it can't
actually remove the noise; it stuffs it elswhere in the spectrum. In playing
with it, I found that the decisions a noise shaping algorithm makes can
result in some very curious sounds.
In the case where the performer(s) are people, and there is cooperation
with a production, this is done by using the audio as a scratch track that
is replaced in post.
Are these Las Vegas fountains?
Bob Morein
(310) 237-6511
Mark
> "Doc" <docsavag...@yahoo.com> wrote in message
well there is another trick...multiband processing
split the audio into 2 tracks
low pass one track at 1 to 3 kHz
highpass the other track at the same cutoff
pass the highpass track through a dynamic range expander
play with the threshold and slope etc so that there is 10 to 20 dB
attenuation when there is no desired HF signal and o dB atten when
there is desired HF signal..
combine the two tracks
This has worked very well for me on voice material. I don't know how
well it will work on music.
The concept is to set the threshold correctly to expand the difference
between the noise and desired HF. If the noise is already louder then
the desired HF, then you may be out of luck.
Mark
Scott Dorsey
>I've got some video with a large fountain that performs shows to
>music. Looking to strip out as much of the "shhhhhh" of the water as
>possible but leaving the music as intact as possible. I've been using
>an app that works with noise samples and am getting a certain degree
>of success but wonder if there's any tricks to refine the results
>further. I'm experimented with varying degrees of eq at various freqs
>but find that a given freq seems to reduce one aspect of the hiss
>while leaving others intact, and also is likely to peel out some of
>the music as well.
Yup.
Most of these systems (at least the CEDAR and Wave ones) break the signal
up into hundreds of bands, and then they gate each one of the bands. Some
of them set the gate parameters based on "training" from a noise sample,
others allow you to set them by hand.
The ones that allow you to set them by hand are a win because it allows
you to do multiple passes with slightly different parameters, and this
means you can do a little more aggressive noise reduction before the
artifacts get annoying.
But really, this is one of the most difficult noise reduction tasks, and
there is no miracle box to do it. All you can do effectively is to get
a clean recording of the music and synch it up.
>I seem to be finding that using the shortest possible sample of noise
>in this case seems to yield better results than using longer samples.
Does your application show you where the gates are set?
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
black...@aol.com
This is an old classic, probably not appropriate and probably out of
date in the digital world:
Run it through a parametric equalizer. Find the approximate frequency
band of the noise you wish to de-emphasize. BOOST the EQ level to
maximum. Sweep the frequency up and down until the noise is loudest.
Then turn the EQ down to CUT. Adjust the bandwidth as narrow as
possible, to get the most noise reduction, but the least effect on the
music. I bet you can find a sweet spot.
david correia
<23271079-ec0c-490d...@21g2000yqj.googlegroups.com>,
Doc <docsa...@yahoo.com> wrote:
If you've got a Mac around and $100:
http://www.bias-inc.com/products/soundsoap2/
David Correia
www.Celebrationsound.com
Arny Krueger
> shows to music. Looking to strip out as much of the
> "shhhhhh" of the water as possible but leaving the music
> as intact as possible. I've been using an app that works
> with noise samples and am getting a certain degree of
> success but wonder if there's any tricks to refine the
> results further. I'm experimented with varying degrees of
> eq at various freqs but find that a given freq seems to
> reduce one aspect of the hiss while leaving others
> intact, and also is likely to peel out some of the music
> as well.
There are two basic approaches.
One is bandpass limiting. The sound of rushing music tends to be broadband
while voices and instruments tend to have narrower bands.
Another is gating. If you only open the signal path when there is a musical
signal, then some of the noise will be removed.
Facilities for doing both bandpass filtering and noise gating are built into
most audio editing software or available as plug ins.
The technique of sampling the noise to synthesize a filter works best when
the noise is not broadband, but has particular spectral content.
Doc
> Are these Las Vegas fountains?
No but same idea. It's a fountain at Epcot.
Doc
>
> well there is another trick...multiband processing
>
> split the audio into 2 tracks
>
> low pass one track at 1 to 3 kHz
>
> highpass the other track at the same cutoff
>
> pass the highpass track through a dynamic range expander
>
> play with the threshold and slope etc so that there is 10 to 20 dB
> attenuation when there is no desired HF signal and o dB atten when
> there is desired HF signal..
>
> combine the two tracks
>
> This has worked very well for me on voice material. I don't know how
> well it will work on music.
>
> The concept is to set the threshold correctly to expand the difference
> between the noise and desired HF. If the noise is already louder then
> the desired HF, then you may be out of luck.
This just sounds interesting but I'm not sure I have all the
nomenclature straight.
What I have to work with is Soundforge 5, which besides the built-in
utilities can also utilize VST plugins. When you say
"..low pass one track at 1 to 3 kHz..highpass the other track at the
same cutoff.."
When you say low pass one track, you mean cut off everything above
some frequency within the 1 to 3 khz range, correct? When you say high
pass, you mean cut off everything below that same freq? The 4-band
paragraphic EQ has a high shelf and low shelf function, is that what
you're referring to? The parametric EQ has a "band pass" eq setting
but it doesn't specify hi/low. It seems the "low shelf" would act as a
high pass and vice-versa.
If I understand what you're saying, this leaves you with say a L
channel that has all low freqs below say 150 khz and one with all high
freqs above 150khz.
When you say
'..pass the highpass track through a dynamic range expander..."
Are you talking about something like the DigitalFishphones Floorfish
on this page?
http://www.digitalfishphones.com/main.php?item=2&subItem=5
It's not entirely clear to me which controls would correspond to
threshold and slope.
When you combine the tracks - i.e. reassign them back to L & R, won't
this result in something odd sounding with a bass only track and the
other with the noise filtered?
hank alrich
http://www.izotope.com/products/audio/RX/
--
ha
shut up and play your guitar
http://www.armadillomusicproductions.com/CarryMeHome.html
http://hankalrich.com/
PStamler
provided you don't overdo them. Likewise in DC SIX -- oops, DC SEVEN.
Go for about 5dB noise reduction and listen very hard for artifacts.
Peace,
Paul
Doc
Repair and playing with it further in SoundForge, I made the
completely accidental observation that applying stereo expansion has
the effect of lessening the remaining noise. Wonder why that is.