Removing unwanted noise

  • Thread starter Thread starter glenroy
  • Start date Start date
Your post seems to be saying "it's rare that you can get perfect results from a noise remover". I agree, but didn't feel like elaborating on something not asked about. You seem to be taking issue with the word "cancel". Call it reverse EQ then. Cool with me. :cool: You're ultimately flipping the sign somewhere and summing sample values, which is an attempt at "cancellation" in my mind. I suppose I could have said "attempt to cancel".

I'm gonna start using "algebraically" to describe more algorithms. Definitely can't have one without. Oh - if you use an FFT at any point, and I can think of several possible uses here, there's trigonometry, too. You could say you trigonometrically remove the analyzed noise sample from the signal. That sounds hardcore - high five. :p
 
By "algebraically" I mean exactly what I said.

(program + noise) - noise = program.

Really high algebra I must admit.

Tim
 
I'm fine :). Don't read my post as snarky - just discussing what you said, and some goofiness at the end. Another smiley fail...I'm trying to get better with the smilies!
 
Tim Gillett said:
By "algebraically" I mean exactly what I said.

(program + noise) - noise = program.

Really high algebra I must admit.

Tim
Click to expand...
Oh, you changed your post after I responded. I like this one better, anyway. The other one seemed kinda hostile :(

(program + noise) - noise = program.
Simplify the algebra by removing program from both sides.
noise - noise = 0.

Hence my response. You can call that a lot of things - the most obvious to me is cancellation. I don't get what your point was aside from the cancellation usually not being perfect, which I agreed with. I don't get this comment about algebra either - I know what you meant, I was just joking around a bit by bringing in other math... I really do think it would a cute marketing line to say "Trigonometrically remove the noise!"
 
TyphoidHippo said:
(program + noise) - noise = program.
Simplify the algebra by removing program from both sides.
noise - noise = 0.

Hence my response. You can call that a lot of things - the most obvious to me is cancellation. I don't get what your point was aside from the cancellation usually not being perfect, which I agreed with. I don't get this comment about algebra either - I know what you meant, I was just joking around a bit by bringing in other math... I really do think it would a cute marketing line to say "Trigonometrically remove the noise!"
Click to expand...

My point is that what some might call "cancellation" is not cancellation. It's just EQ. That's all noise reduction software does, except that it's dynamically changing EQ.
What you call cancellation I call EQ.

Lets say you have an unwanted 400hz tone all through your track. So you use a high Q band reject filter and reduce the tone by 20db. Unfortunately you have now reduced all program at 400hz by 20db as well! Have you separated the unwanted 400hz tone from the 400hz program material? No. Not even a tiny bit.

OTOH if you were able to feed back into the program a 400hz tone very similar to the one you want to eliminate, and make it 180deg in antiphase to the original tone, you will have reduced the tone but also not touched the program material at all. Not even a tiny bit.

That is cancellation, and normally what we mean by the term cancellation in audio terms..

The second method is much harder to implement but if and when you can, it actually works. That's why we observe the "three to one" rule with mic placement. We observe it because unfortunately there, "cancellation" also "really works", especially when we dont want it to!

The central problem with noise reduction is differentiation. Separating the noise from the program. That's why I bracketed them together as in: (program + noise). Normally they're stuck together like glue and cant be torn apart.

It's the easiest thing to buy a mixer and mix two signals together. Why is it so much harder to go out and buy an "unmixer", or a "separator" if you like, and unmix what is mixed together?

Tim
 
I was rather impressed with the Sonnox Denoiser included with Wavelab. I was able to take the edge off the noise in my older recordings without compromising the recording at all.

Of course, getting all the noise out will almost always take off your high frequencies, therefore making it best to re-record if you can. Boost your frequencies with EQ and the noise comes right back. :( Sony's restoration tools might be worth checking out as well.
 
I was listening to some of my earlier recordings, some that I made when I first started recording with decent software. However, the computer I was using had one of those jet-engine CPU cooling fans that always made it onto anything that was recorded with a mic. It was a real pain in the fundament for a long time. I had to move the mic to another room and close the door to get anything useful.

But I listened to some of those first recordings and although the performance was acceptable to me, the background noise of the computer fan was just so irritating and unprofessional. Just recently I used iZotope RX to sample the noise at a quiet point at the beginning or end of the song and let iZotope take a profile from that. The result was outstanding. In some cases the noise was removed completely, in others down to an acceptable level. As I get better with iZotope, I am sure such corrections will be a lot better. My most recent recordings don't need it as my technique and gear have improved a lot but I still liked some of those older recordings that I don't particularly want to do again.
 
TyphoidHippo said:
Your post seems to be saying "it's rare that you can get perfect results from a noise remover". I agree, but didn't feel like elaborating on something not asked about. You seem to be taking issue with the word "cancel". Call it reverse EQ then. Cool with me. :cool: You're ultimately flipping the sign somewhere and summing sample values, which is an attempt at "cancellation" in my mind. I suppose I could have said "attempt to cancel".

I'm gonna start using "algebraically" to describe more algorithms. Definitely can't have one without. Oh - if you use an FFT at any point, and I can think of several possible uses here, there's trigonometry, too. You could say you trigonometrically remove the analyzed noise sample from the signal. That sounds hardcore - high five. :p
Click to expand...

Don't confuse the issue with facts please?
 
GuitarLegend said:
I was listening to some of my earlier recordings, some that I made when I first started recording with decent software. However, the computer I was using had one of those jet-engine CPU cooling fans that always made it onto anything that was recorded with a mic. It was a real pain in the fundament for a long time. I had to move the mic to another room and close the door to get anything useful.

But I listened to some of those first recordings and although the performance was acceptable to me, the background noise of the computer fan was just so irritating and unprofessional. Just recently I used iZotope RX to sample the noise at a quiet point at the beginning or end of the song and let iZotope take a profile from that. The result was outstanding. In some cases the noise was removed completely, in others down to an acceptable level. As I get better with iZotope, I am sure such corrections will be a lot better. My most recent recordings don't need it as my technique and gear have improved a lot but I still liked some of those older recordings that I don't particularly want to do again.
Click to expand...

I get your point, but I just use a gate to get rid of low level sounds like computer fans and kids puking in another room.

I could sell you something else but a gate might do the trick!
DW
 
YellowDwarf said:
I get your point, but I just use a gate to get rid of low level sounds like computer fans and kids puking in another room.

I could sell you something else but a gate might do the trick!
DW
Click to expand...

Thanks for the offer. I use a noise gate where it is appropriate like where the alleged "silence" is. But when the noise is mixed in with signal that you want, a noise gate doesn't do the trick. You still have noise under the audio that you want to keep. iZotope RX handled it beautifully.
 
YellowDwarf said:
Ah, I see. The noise is at the same volume as the signal.
Click to expand...

Not that you could hear it when the song started but the leadin, which nowadays I just mute out, carried the distinct sound of guitar buzz or computer fan scream. The "silence" at the trailoff of the song also carried that noise as did some quiet phrases within some songs. I profiled the noise where there was no other signal in the leadin which in my ignorance I had not muted out and it was enough for iZotope to work with. The result was excellent throughout the recording
 
The problem with noise sampling or fingerprinting is that the freqs associated with the noise are filtered out of the recording so they are removed from the music as well as the non music. Careful use of that process is needed. Waverepair has the same process - & the sample called a fingerprint. It has some clever things like non destrustively removing everything BUT the sample's freq. so that you can hear how much of the music is removed before dtermining to remove it. the level of removal is variable too PLUS an option to only remove it at quieter parts of the song.
I've done a fair bit of those things with my older tape recordings and some early fan noise too. Often quite successfully BUT I learnt to save a copy of the original before removing and saving so that I had the option to go back if the results, a few days later, seemed less musical than I'd thought.
I had an ME work on some tracks that had a bit of noise in them and we discussed denoising but his position was that topping & tailing the track as well as being careful with levels in the mix often had a more musical result that the filters.
 
You must log in or register to reply here.
Back
Top