A better noise reduction tool?

johnmeyer wrote on 4/11/2007, 5:34 AM

It does sound interesting. Some of the problems it claims to fix (e.g., coughs, chair squeaks, etc.) can be fixed in the time domain in SF using the various "replace" functions. This works well during music with a sustained note or with speech, if the presenter is paused between words, but can create distracting echos or repeats in other situations. Therefore, a better solution would be welcome.

They claim that "every spectral modification is executed in linear-phase domain." As an electrical engineer who studied Fourier functions in college, I am familiar with the time to frequency domain transformations. I am not familiar with the phase domain so it's tough to know exactly what is going on and whether it would be successful. It is certainly true that the trick to good noise reduction is figuring out whether to perform the cleanup function on the frequency (i.e., spectral) or time representation of the audio.

In reading further, it sounds like perhaps they are doing something simultaneously in both domains, which would be a pretty interesting technical challenge. Here's their description: "and displays the result as a 3D spectrogram with time on the horizontal axis, frequency on the vertical axis and amplitude of the spectral components color-coded ..."

I note that this product costs 2,490 euros, so this is a BIG purchase. You would think they would have before/after demo samples all over their site, especially since their motto is "experience the difference." I could find none. However, they do have free demo downloads.

I would be very interested if anyone has the time (I don't) to test this or has already used it. It's probably outside my price range, but I sometimes get clients that have this kind of money who might underwrite my purchase.

[Edit] OK, I did find some samples, but for their SoundLaundry product, which seems to different than the product discussed here, and actually almost the same as the SF NR plugin. Here's the link to the page where you can hear what sounds like the introduction to Smetna's Moldau. The before/after sounds exactly like what I would get with the SF NR plugin (actually, I think NR does a better job).

SoundLaundry Samples

I did a Google search of blogs, and came up with this rather complete review:

Review

JJKizak wrote on 4/11/2007, 6:06 AM

I must be from another planet. I always add coughs and chair squeaks and wind noise to the old home movies without sound tracks to make them sound real.
JJK

johnmeyer wrote on 4/11/2007, 6:26 AM

I must be from another planet. I always add coughs and chair squeaks and wind noise to the old home movies without sound tracks to make them sound real.

And you probably add that horrible faux dust and scratch fX to your video to make it look like Super8! :)

craftech wrote on 4/11/2007, 6:37 AM

What is needed is a NR tool that can remove varaible frequency noise effectively. Most of those like NR2 are effective at removing constant frequency noise. Example of variable frequency noise that is impossible to remove: "crickets" when you film something like an outdoor play.

John

Coursedesign wrote on 4/11/2007, 6:51 AM

Crickets are definitely the worst! I spent quite a few hours before giving up on a recording where I had no control of either the crickets or the subject.

Adobe had a very impressive noise removal demo on their tour last year. Audition can show a two-dimensional noise spectrogram, and you can go in with an eraser tool to remove even complex noises during complex music say.

The spectrogram was actually in two dimensions + color, so it was easy to see what belonged there.

This at least sounds like the 2,490 euro product described.

johnmeyer wrote on 4/11/2007, 7:46 AM

I think this tool might be able to do something with the crickets. In searching for audio before/after examples, I found one discussion on a forum somewhere where a user claimed to have been able to fix one singer's flat entrance in a duet. Apparently the singer started the duet off-key and then quickly adjusted. Her tonality was sufficiently different from the other singer that he was able to isolate her and pitch adjust the few offending seconds. Sounds totally impossible to me, but that was the claim. I read a couple of other equally implausible stories, but they were all posted on what seemed like a fairly reputable forum, so maybe it's for real.

farss wrote on 4/11/2007, 8:18 AM

Maybe it is possible, NR2 is pretty good and I've had good results by doing the hard yards modifying noise prints etc. It could sure use some polish though. But I don't think it's overly smart, it's really no more than a massive bank of notch filters with a way to dial in the attenuation of each band from a noise sample by Fourier analysis.
Those hiQ filters could indeed be introducing phase errors which certainly wouldn't help, plus reNOVAtor seems to be able to understand harmonic relationships.

I guess I should submit a trial request, what's to loose. I'm just a bit leary of some of the higher end audio tools. This one is pitched at the top end of the market where another $5K would be neither here or there. Cedar is a good example, the local agent loved to tell me how "magic" it was and how anything cheaper simply must be trash. When that didn't wash he insisted SoFo had "stolen" the IP from Cedar so if I was a half decent bloke I'd fork out $14K anyway, yeah right!

Speaking of Fourier analysis, over drinks an old workmate was telling me about Lake and his work on long line fourier analysis. This is the stuff used in the Lake audio processor which the sound reinforcement guys now swear by to tame huge arenas. Maybe some of this tech has found it's way into other areas, it's all a bit over my head.

Bob.

RBartlett wrote on 4/11/2007, 12:15 PM

<johnmeyer quote from earlier>

They claim that "every spectral modification is executed in linear-phase domain." As an electrical engineer who studied Fourier functions in college, I am familiar with the time to frequency domain transformations. I am not familiar with the phase domain so it's tough to know exactly what is going on and whether it would be successful. It is certainly true that the trick to good noise reduction is figuring out whether to perform the cleanup function on the frequency (i.e., spectral) or time representation of the audio.

</quote>

Same discipline myself John, and you covered all you needed I'm sure. Phase is a comparitive measure of frequency and it's incidence and alteration over time or range. I'd wager you'll remember phasor diagrams. You may also recall finite impule response filters, there your attention would have been drawn to the benefits of signal processing in that modifications can be designed to occur equally in "delay" over all frequencies.

What is linear-phase related to with a filter? - See here

I've forgotten more than I care to say. I found some aspects of time, frequency, phase and transmission line theory as abstract as the concept of whether light is a particle or an excitation of magnetic/electric energy that is measured as a frequency. Apparently it is all these things.

filmy wrote on 4/11/2007, 12:50 PM

>>>Adobe had a very impressive noise removal demo on their tour last year. Audition can show a two-dimensional noise spectrogram, and you can go in with an eraser tool to remove even complex noises during complex music say.<<<

Soundbooth has this feature. At the CS3 launc they bascialy said they wanted to build an app so someone with no knowledge of audio could just "click a button" and it would do everthing the "right way" behind the scenes. I have payed with Soundbooth and it is good, but I am so used to going into sound and actually working with it is is just too wierd to click a single button. As for the eraser tool - it is clever but it still needs you to somewhat isolate what the sound is. Variable sounds would need more work.

To go back to the orginal question - I use the Waves plug-ins. However crickets are still a pain no matter what tool you use. I would be happy if someone came out with a one button cricket remover that worked though.

epirb wrote on 4/11/2007, 1:45 PM

I have just started using Sonys NR tool and am still learning, however, I find that Virtos' Noise wizard works very well for many of my needs ie removing air cond hiss etc. In its dynamic mode applied at the track level can work wonders.
Albeit not a hi end Noise reduction like the one you mentioned for its price it does suprisingly well.

Opampman wrote on 4/11/2007, 3:23 PM

For those not familiar with it, let me suggest another good option for noise reduction and audio processing - DC6 from Diamond Cut Productions. This software was originally designed for removing background noise, pops, clicks and the like from old audio disc but has evolved into much more. It will open your AVI files with video and audio directly from Vegas with a right click and has incredible filters, noise reduction and audio processing tools with every parameter adjustable. You want your audio to sound like it went through a tube preamp - DC6 does it and you select the tube - 12AX7, etc. I use both Sound Forge and DC6. But, the DC6 noise reduction works great and the various equalizers, filters, compressors and expanders are extremely flexible

http://www.diamondcut.com/

FuTz wrote on 4/11/2007, 7:14 PM

Maybe this?
http://www.sydec.be/Products/Details/ID/7a109a03-3a40-4426-b735-ee911916cd97/reNOVAtor%e2%84%a2/

...Renovator too but by Sydek : is that the same thing?

Edward wrote on 4/12/2007, 4:57 PM

Hey Bob,
I was at last year's NAB at the Adobe Audition demonstration. I saw the guy remove a squeak of a door, and nothing else was efected. He did this with photoshop like tools (lasso). He then pasted that selected sample everywhere in the sound file. Amazing stuff.

Sorry, I've been on an Adobe craze and I ain't tryin' to push this software. I was hoping that SF could do some of the things Audition could. I've been using SF since version 4. Never saw anything like what Audition could do.

Ed.

johnmeyer wrote on 4/12/2007, 5:13 PM

Phase is a comparitive measure of frequency and it's incidence and alteration over time or range. I'd wager you'll remember phasor diagrams. You may also recall finite impulse response filters, there your attention would have been drawn to the benefits of signal processing in that modifications can be designed to occur equally in "delay" over all frequencies

Yeah, phase is basically "relative frequency," meaning that the number of cycles per second is the same, but offset by some amount in time. However, the math does get different. Back when I worked at Hewlett-Packard's microwave test & measurement division in Palo Alto, we briefly studied frequency shift keying vs. phase shift keying, two ways to modulate data onto a carrier. They were different, but only subtly.

As for impulse response, that was the cornerstone of EE101 at my school, where you would apply (theoretically) an infinitely short duration burst of energy to a circuit and then calculate the response. This provided the natural response of the circuit, and I can remember spending lots of time with a slide rule (yes, I am that old) calculating pole-zero diagrams. It's been thirty-five years since I studied all that and I can still remember the basic ideas, although the details are a little rusty (as I found out when I tried to help my son factor a polynomial a few months ago).

Anyway, the stuff I read about this particular noise reduction tool makes it sound like FM. No, not Frequency Modulation, but what a salty old engineering manager used to call F****** Magic.

Too bad it is so much money or I'd go for it in a heartbeat. You think the Madison folks are ever going to truly update NR? This would be a lot of fun for some UofW pHd.