Waves Restoration, Wave Arts Master Restoration Suite, iZotope RX?

[ndogg]

What's your favourite plug-in (ProTools on Intel Mac) for production de-noise tasks? Mostly for removing ambient background noise in location audio.

Maybe I should start a poll?

Anyone who has had the chance to compare any of these directly, your input would be greatly appreciated.

Thanks,
n

[Whitecat]

Originally Posted by ndogg

What's your favourite plug-in (ProTools on Intel Mac) for production de-noise tasks? Mostly for removing ambient background noise in location audio.

Maybe I should start a poll?

Anyone who has had the chance to compare any of these directly, your input would be greatly appreciated.

Thanks,
n

The best is iZotope RX. iZotope RX - Complete Audio Restoration: Declipping, Declicker, Hum Removal, Denoiser, Spectral Repair, Restore, Remaster, Download

[dogandpony]

Of the three I have had the best success with Izotope. john.

[danijel]

WaveArts NR is extremely low-latency and low-cpu, and can be used as insert on every dialogue channel. And the noiseprint can be automated, although not very conveniently.

[Rick Sanchez]

I have had good success with the Izotope and fair results with Waves

[wolfdavis]

It really depends on what you need to reduce.

Hum and buzz.... Waves.
Pops and clicks... Waves.
Random oddities in the recording.... iZotope.
Missing segments or errors in recording... iZotope.
60-120Hz hum. Any ol' EQ unless it's a bass or LF instrument then re-record.

Hope that helps.
Rocky Mountain Recorders - The Professional Recording Studio in Denver, CO
ROCK.

[TVPostSound]

Izotope RX has a very low price tag, even though I have the Waves restoration, you can't beat the RX's de-clipping function.

[marty lester]

i just used the hell out of Izotope on a documentary that largely
took place in a tattoo shop. while it didn't remove the noise
completely by any means, there were some useful results
by using the hum remover as a several-band notch filter.

i did, however, remove a squeal/squeak type sound from an
audio track using (Izotope) Spectral Repair. unbelievable. the editor
was amazed.

for general noisy environments, RX does ok but frankly, nothing
i've tried does it really well. i may have to try the Waves someday.

good luck,
marty.

[kbshowle]

I've had the best results with WaveArts with for any tape restoration. I preferred it over Izotope, Waves, Sonic NoNoise, any Sony, and Digi DINR.

Has anybody had experience with CubeTec restoration.

[Branko]

I have best results with WaveArts MR Noise. I don't find it difficult to automate the curve, if you use a "copy/paste settings" procedure. As Danijel pointed out, low latency is one of the advantages. Waves X-noise produces artifacts easily, while NoNoise's latency (over 4000 samples) makes it useless except for off-line processing.

[Airon]

On tonal noise, Izotope denoising absolutly rules. I've cleaned a whole bunch of scenes with lamp buzzes and whines this way. You can denoise the tonal and broadband noise independently.

Waves decrackle is still a nice one for wireless rustlers.

If I could, I'd keep all the tools handy. For now, the RX bundle is a brilliant package to own, if you can only afford one right now. There's plenty of others, but this one does its job on more occasions than the others for me.

[evangilmore]

Izotope Rx works incredibly well. In both standalone and plugin mode, you can't beat it for removing backround noise, hiss, hum, etc. Definitely my favorite tool for this kind of restoration.

[Dragomir]

How much cpu hungy is the VST RX compared to WaveArts MSNoise?

[Fredo]

If you all think that Izotope and Waves are so brilliant, then you really should check out the Algorithmix stuff. Blows anything else away.

Fredo
Temple of Tune
Belgium

[Whitecat]

Originally Posted by Fredo

If you all think that Izotope and Waves are so brilliant, then you really should check out the Algorithmix stuff. Blows anything else away.

Fredo
Temple of Tune
Belgium

Having a real hard time to get it to work under OSX unfortunately.

[DrDeltaM]

Originally Posted by ndogg

What's your favourite plug-in (ProTools on Intel Mac) for production de-noise tasks? Mostly for removing ambient background noise in location audio.

Cedar DNS. Actually a hardware unit tho, but comes with a control plugin for ProTools integration (DNS2000 & DNS3000).

Removing background noise in location audio was what it was primarily designed for Works fast and easy with great results.

[DrDeltaM]

Originally Posted by evangilmore

Izotope Rx works incredibly well. In both standalone and plugin mode, you can't beat it for removing backround noise, hiss, hum, etc. Definitely my favorite tool for this kind of restoration.

I dare to say that Cedar blows Izotope Rx away Of course we're talking a different price-level too here, but stating that Izotope Rx can't be beat is not really correct

[Alexey Lukin]

Here's a comparison of noise reduction systems on a single short speech file with a broadband noise and hum. It's hard to draw conclusions about quality from a single example, but you can at least estimate what artifacts to expect from different systems.

[danijel]

Originally Posted by Branko

I have best results with WaveArts MR Noise. I don't find it difficult to automate the curve, if you use a "copy/paste settings" procedure. As Danijel pointed out, low latency is one of the advantages. Waves X-noise produces artifacts easily, while NoNoise's latency (over 4000 samples) makes it useless except for off-line processing.

Branko, since you pointed me towards WaveArts MR Noise, I had a real tough test drive for it! I was helping out some friends that do a TV show (dramatic) that is shot with three RED cams simultaneous! The smaller the space they shoot in, the bigger the noise. Needles to say, boom mic is what I mute first, and then checkerboard the lavs.

I did the first episode using ZNoise and I had to clean EVERY clip off-line, and it took me double the time to edit/mix dialogue than I thought it will.
On the next episode, I tried WaveArts on every dialogue channel (as you suggested), but I had it online WHILE editing dialogue, and it felt almost like editing normal dialogue, except that I had to be careful not to change gain BEFORE dialogue enters the MR Noise (so, no clip gain). I made a 'RED CAM NOISE' noiseprint (although, obviously, it is not always the same, but I had no time to automate it for every scene & angle), and automated threshold only.

So, I can't say if it's better or worse than any other NR, just that it enables much faster workflow, when time is critical. And all that I said applies to broadband noise only.

[jahtao]

Originally Posted by danijel

On the next episode, I tried WaveArts on every dialogue channel (as you suggested), but I had it online WHILE editing dialogue, and it felt almost like editing normal dialogue, except that I had to be careful not to change gain BEFORE dialogue enters the MR Noise (so, no clip gain). I made a 'RED CAM NOISE' noiseprint (although, obviously, it is not always the same, but I had no time to automate it for every scene & angle), and automated threshold only.

V Cool!!

[philper]

Originally Posted by danijel

Branko, since you pointed me towards WaveArts MR Noise, I had a real tough test drive for it! I was helping out some friends that do a TV show (dramatic) that is shot with three RED cams simultaneous! The smaller the space they shoot in, the bigger the noise. Needles to say, boom mic is what I mute first, and then checkerboard the lavs.

I did the first episode using ZNoise and I had to clean EVERY clip off-line, and it took me double the time to edit/mix dialogue than I thought it will.
On the next episode, I tried WaveArts on every dialogue channel (as you suggested), but I had it online WHILE editing dialogue, and it felt almost like editing normal dialogue, except that I had to be careful not to change gain BEFORE dialogue enters the MR Noise (so, no clip gain). I made a 'RED CAM NOISE' noiseprint (although, obviously, it is not always the same, but I had no time to automate it for every scene & angle), and automated threshold only.

So, I can't say if it's better or worse than any other NR, just that it enables much faster workflow, when time is critical. And all that I said applies to broadband noise only.

How many dx channels did you have the WaveArts plug working in at once? Did you "feel" the CPU load of doing this? This approach is very interesting to me if the load can be managed.

Philip Perkins

[Dragomir]

I made an A/B test in Samplitude: RX Denosier vs. MR Noise (realtime, of course) on a hissy Foley track. On my opinion, RX sounds better, but the CPU load is 24-26% compared to the 6-8% of the MR Noise. Algorithmics is very tempting, but for the price tag , I won't even try the demo.

[danijel]

Originally Posted by philper

How many dx channels did you have the WaveArts plug working in at once? Did you "feel" the CPU load of doing this? This approach is very interesting to me if the load can be managed.

Philip Perkins

4 channels. I didn't "feel" them, probably because it was a fairly simple project with only a handful of SFX and BG channels and one reverb instance.

Each MR Noise takes 5% CPU here (but I'm at a point of going for new hardware). It has 640 samples of latency.

I still have 15 days of trial, so I'll look further into automating the noiseprint more conveniently.

[Branko]

In my latest project, I had 8 MR Noise plugins on my dialog tracks. My daw is ProTools HD, and the cpu I'm using is Mac Pro Dual 2.0 (4 processors in total) I use only 3 processors for Rtas plugins and allow the load up to 95 %. Everything worked flawlessly.
Sorry to insist, but the right noise print is essential for a good de-noising.

I developed a very simple working method:
1. suspend automation
2. select a small piece of audio containing noise only
3. set your MR Noise learn time to 200-500ms (200 is usually fine)
4. run it to learn the noise - MR Noise automatically drops out of the learn mode after predefined time.
5. select the whole clip and tweak until you get rid of the noise. You may tweak the curve, separate reduction values for each band etc.
6. When you're happy, stop and copy the parameters (shift+command+C)
7. position the playhead/cursor before the clip
8. in "latch" mode (with "latch prime on stop" enabled) paste settings (shift+command+V) if you don't like latch prime, use "write"mode.
9. hit play and enjoy!

Although it may seem a bit complicated (9 steps to follow), it is actually a very quick way to automate the whole track.

I have made a preset with all parameters set to defaults I like, reduction set to 9-12dB, with the noise curve at - infinity, and I keep the plugin always in. Bypassing may change the latency value and cause phase shifts and echo in multichannel recordings.

Just one more word: we're talking about a very cheap plugin, so I cannot admit any comparison to Algoritmix, Cedar & co. Btw, on some program, MR works much better than Cedar.
Another thing I find worth commenting: I've seen a comparison between various hard and software denoisers, and I read there that many of tested products were not perfect because they failed to reduce hum or other static, single-band noises. In my opinion, it is an obvious case of wrong use of denoiser. All denoisers work much better if you remove narrow-band, constant frequency tones witn a decent notch filter before the denoiser.
My denoising chain consists of Equium followed by MR Noise. If I have time, or when I need something different, or simply just want to impress the client, I may plug in the Dolby Orange Box.
Branko

[ndogg]

Thanks for all the input, everyone.

We ended up demoing and buying the iZotope RX Advanced stuff... it's a BIG improvement over the Digi DINR plug-in we have. And the spectral repair stuff is really impressive.

I was aware of the Cedar stuff, but it's too pricey right now and I'd rather stay in the box. Maybe someday...

Thanks again.

Cheers,
n

[philper]

I compared MRNoise and Izotope RX on a job yesterday. I was able to get MRNoise pretty close to what Izotope can do on many sorts of BG noise (in a doc), and being able to run it as a live plug in might push me to try even harder to make it work for me.

thanks

Philip Perkins

[James Meeker]

Izotope RX is like magic.

[Branko]

I took some time to listen carefully to noise reduction "shootout" done by Alexey Lukin.

First of all, some of the tested systems (plugins) were used in such a way, that they couldn't produce any satisfying results.

Let's begin with Cedar DNS 1000: The threshold is set too low, and the reduction is set too high - on every band. Obviously, it will not produce any useful result. On HF, it was not necessary to reduce more than 6-9 dB, and the authors went all the way down.
TC Backdrop - set up pretty good, just a slight "underwater" effect on breaths
Sonic foundry - big problems with breath, yet the authors report only "incomplete suppression of hum"
MR noise - too much reduction, absolutely wrong noise curve, threshold too high.
IZotope - heavy noise reduction, serious artifacts on breaths.

Second, in post-production, applying NR to a studio recording is a rare case. The real challenge for a NR system (or plugin) is a noisy production track.

In most cases, authors didn't pay attention to artifacts NR plugins created on breaths, which are the most critical and most revealing.
A general conclusion is that, while trying to kill the hum, authors set the parameters too high and ended up destroying the voice. A good engineering practice has shown that hum and other static, narrow-band tones, require different tools, in most cases a good set of notch filters.

[Alexey Lukin]

Originally Posted by Branko

I took some time to listen carefully to noise reduction "shootout" done by Alexey Lukin.
First of all, some of the tested systems (plugins) were used in such a way, that they couldn't produce any satisfying results.

Thanks for your comments, but I'd appreciate more constructive approach. Those examples were produced by different people instructed to achieve around 12 dB of noise reduction and maximize the quality of every tested system. Are you interested in improving the shootout?

If you feel that some systems have been misconfigured - you are welcome to produce better examples, and I'll gladly place them on the web page.

Originally Posted by Branko

Let's begin with Cedar DNS 1000: The threshold is set too low, and the reduction is set too high - on every band. Obviously, it will not produce any useful result. On HF, it was not necessary to reduce more than 6-9 dB, and the authors went all the way down.

The threshold has been set to a rather reasonable value, but I also have the version with a lower threshold for this unit. It has a different kind of noticeable artifacts: noise gates are triggered by speech too often and the noise modulation is clearly audible. Also, the hum is barely suppressed.
The condition of the test has been to reduce all noise by about 12 dB (and reduce hum even further, if possible). This makes it possible to hear the musical noise artifacts created by most systems. At 6 dB they are barely audible.

Originally Posted by Branko

MR noise - too much reduction, absolutely wrong noise curve, threshold too high.

Please produce a better sample if you can. The amount of reduction should be 12 dB, or even more on hum.

Originally Posted by Branko

Sonic foundry - big problems with breath, yet the authors report only "incomplete suppression of hum".
IZotope - heavy noise reduction, serious artifacts on breaths.
In most cases, authors didn't pay attention to artifacts NR plugins created on breaths, which are the most critical and most revealing.

You seem to hear something that I didn't. Can you explain the kind of artifacts on breath that you are hearing? Which segment of the file is most affected? Are there examples of proper noise reduction that don't create such artifacts?

Originally Posted by Branko

A general conclusion is that, while trying to kill the hum, authors set the parameters too high and ended up destroying the voice. A good engineering practice has shown that hum and other static, narrow-band tones, require different tools, in most cases a good set of notch filters.

My practice has shown that in many cases a properly configured denoiser works more accurately to the signal than a set of notch filters. The spectrum of the hum in this example is very broad, which makes notch filtering impossible.

[Branko]

My practice has shown that in many cases a properly configured denoiser works more accurately to the signal than a set of notch filters. The spectrum of the hum in this example is very broad, which makes notch filtering impossible.

As I said, my practice has shown exactly the opposite. I will try to produce some samples using notch filters and denoisers, if that's permitted.

As for the artifacts, they can be easily heard every time the VO artist breathes in.