Forums

Removing White Noise

Started by Duane September 22, 2005
Hey everyone,

I have an audio application, and I'm trying to do some pre/post
processing to improve the sound quality.  I put in an adaptive filter
to try and remove structured noise, but I'd like to remove white/pink
noise from the signal somehow.. I've been looking at wavelet
techniques, or whatever else I can find.

Can anyone recommend a good way to reduce white noise from audio/

Also, in general, any suggestions on how to improve voice quality?

Thanks,
Duane

Duane wrote:
> Hey everyone, > > I have an audio application, and I'm trying to do some pre/post > processing to improve the sound quality. I put in an adaptive filter > to try and remove structured noise, but I'd like to remove white/pink > noise from the signal somehow.. I've been looking at wavelet > techniques, or whatever else I can find. > > Can anyone recommend a good way to reduce white noise from audio/
In the late 40's, H.H.Smith developed a dynamic noise suppressor for shellac records. It divided the entire band into sub-bands, and suppressed those sub-bands, including any noise in them, with small energy. Obviously, digital processing allows this idea to be improved in many ways. I understand that the best noise-reduction systems use methods descended from Herman's.
> Also, in general, any suggestions on how to improve voice quality?
If the poor quality is caused by poor frequency response, equalizers can help. (They can improve balance, but they can't restore missing frequencies.) If that's not the vase, singing lessons might be in order. Jerry http://65.219.61.150/forums/printthread.php?t=19694 -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Jerry Avins wrote:
> Duane wrote: > > Also, in general, any suggestions on how to improve voice quality? > > If the poor quality is caused by poor frequency response, equalizers can > help. (They can improve balance, but they can't restore missing > frequencies.) If that's not the vase, singing lessons might be in order.
Don't they make such things as rack-mount autotuners, which can re-tune a performers singing voice back on key? There are also vocoders, which can make someones voice sound like a robot, a distinct improvement for some "singers". I suppose an additive vocoder could even be used to restore some of the missing frequencies, if you have some external knowledge (a newer recording of a similar song by the same artist, for instance) of what might be missing on some particular vowel tone. IMHO. YMMV. -- rhn A.T nicholson d.O.t C-o-M
"Duane" <duanestorey@excite.com> writes:

> Hey everyone, > [...] > I'd like to remove white/pink > noise from the signal somehow.
Hey Duane, Yeah, I've got an algorithm that can do that. Let me finish up on this perpetual motion machine and I'll send it to you. -- % Randy Yates % "Ticket to the moon, flight leaves here today %% Fuquay-Varina, NC % from Satellite 2" %%% 919-577-9882 % 'Ticket To The Moon' %%%% <yates@ieee.org> % *Time*, Electric Light Orchestra http://home.earthlink.net/~yatescr
"Randy Yates" <yates@ieee.org> wrote in message 
news:y85ovbcf.fsf@ieee.org...
> "Duane" <duanestorey@excite.com> writes: > >> Hey everyone, >> [...] >> I'd like to remove white/pink >> noise from the signal somehow. > > Hey Duane, > > Yeah, I've got an algorithm that can do that. Let me > finish up > on this perpetual motion machine and I'll send it to you.
Using PowerPoint engineering, it's really very simple. First, you subtract the desired signal from the corrupted signal leaving white/pink noise (other colors are more difficult). Next, you subtract the noise component from the corrupted signal leaving the desired signal which you then subtract from the corrupted signal leaving white/pink noise. Then you subtract the noise component from the corrupted signal leaving the desired signal... Application note: When coupled to a perpetual motion machine, this produces a noiseless prime mover suitable for driving almost anything that requires no power.
"Randy Yates" <yates@ieee.org> wrote in message news:y85ovbcf.fsf@ieee.org...
> "Duane" <duanestorey@excite.com> writes: > >> Hey everyone, >> [...] >> I'd like to remove white/pink >> noise from the signal somehow. > > Hey Duane, > > Yeah, I've got an algorithm that can do that. Let me finish up > on this perpetual motion machine and I'll send it to you.
Actually, if the algorithm can "hear" a snapshot of noise only and the noise level is relatively constant, removing broadband noise is possible. Lots of teleconferencing products have noise removal algorithms for HVAC, laptop fans, etc. and they work even on broadband "air movement" type noises. I've heard them, and they work (though sometimes they generate objectionable artifacts when SNR is poor to begin with). So it's not quite as hard as a perpetual motion machine. :-)
"Randy Yates" schrieb
> > > [...] > > I'd like to remove white/pink > > noise from the signal somehow. > > Yeah, I've got an algorithm that can do that. Let me > finish up on this perpetual motion machine and I'll > send it to you. >
:-) But remember that the famous excuse "I have an explanation, but the margin is too small to hold it" does not work here on the internet. SCNR Martin
rhnlogic@yahoo.com wrote:
> Jerry Avins wrote: > >>Duane wrote: >> >>>Also, in general, any suggestions on how to improve voice quality? >> >>If the poor quality is caused by poor frequency response, equalizers can >>help. (They can improve balance, but they can't restore missing >>frequencies.) If that's not the case, singing lessons might be in order. > > > Don't they make such things as rack-mount autotuners, which can re-tune > a performers singing voice back on key? There are also vocoders, which > can make someones voice sound like a robot, a distinct improvement for > some "singers". I suppose an additive vocoder could even be used to > restore some of the missing frequencies, if you have some external > knowledge (a newer recording of a similar song by the same artist, > for instance) of what might be missing on some particular vowel tone.
You must be right. They frequently do that on TV shows. Jerry -- Engineering is the art of making what you want from things you can get. &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
Randy Yates wrote:
> "Duane" <duanestorey@excite.com> writes: > > >>Hey everyone, >>[...] >>I'd like to remove white/pink >>noise from the signal somehow. > > > Hey Duane, > > Yeah, I've got an algorithm that can do that. Let me finish up > on this perpetual motion machine and I'll send it to you.
Poor Herman (I knew him*, Horatio) might take issue with you, as I describe (with very broad brush strokes) yesterday. Jerry _______________________________________ * I politely chewed him out once. The push-pull output stage of his power amplifier used the filament of the 12.6-volt low-level tube as its load in order to minimize hum pickup. The usual 200 microfarad bypass 'lytic was there. Before I knew any of this, I pulled the input tube to test its Gm -- it was OK -- and replaced it. The filament blew out with the surge from the cap, and it took me a long time to find that. (The flash doesn't show up with a metal tube; 12SJ7.) After all, I knew the tube was good because I had just tested it. Once I figured out it was dead, I plugged in a new one and promptly blew that. A warning label on the chassis would have been nice! -- Engineering is the art of making what you want from things you can get. &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
"Duane" <duanestorey@excite.com> wrote in message 
news:1127408349.860752.243720@g44g2000cwa.googlegroups.com...
> Hey everyone, > > I have an audio application, and I'm trying to do some pre/post > processing to improve the sound quality. I put in an adaptive filter > to try and remove structured noise, but I'd like to remove white/pink > noise from the signal somehow.. I've been looking at wavelet > techniques, or whatever else I can find. > > Can anyone recommend a good way to reduce white noise from audio/ > > Also, in general, any suggestions on how to improve voice quality?
Duane, I've not done it .... but I'd conceptually start with an Adaptive Line Enhancer (ALE) design that removes white noise by passing the entire signal through an adaptive filter that bandpasses the tones. I'd then ponder what changes would be needed for an audio application. The success of an ALE depends on long term stability of the desired signal. Since you don't have that situation exactly in audio then one might imagine that the time constants would be shorter, the passbands wider, etc. Maybe this hits a dead end and maybe not. I think it boils down to the method that Jerry mentioned - perhaps with a different implementation approach. Cool Edit has / had? a filter to remove scratches and pops from phonograph records - maybe that would work for you. I think it hurts the high frequencies though... maybe not... Fred