noise reduction with isotope rx

[cinetj]

Hi everyone,

Trying for the first time Izotope RX. I'm using the denoise tool for cleaning a poor recorded audio. Although I can erase completely the noise in the silent parts I can still hear a little of the noise when the voice kicks in. The treshold seems to solve the issue but then I start hearing artifacts more...

What shoul I be tweaking them?

[danijel]

Originally Posted by cinetj

The treshold seems to solve the issue but then I start hearing artifacts more...

Did you try the "smoothing" slider for reducing artefacts? That's what separates RX from other NRs.

[Uncle Bob]

It's sometimes better to reduce the noise a little bit at a time instead of all at once - one pass for the rumble, one pass for the hiss, one pass for the hum, etc.

[beanface]

Originally Posted by Uncle Bob

It's sometimes better to reduce the noise a little bit at a time instead of all at once - one pass for the rumble, one pass for the hiss, one pass for the hum, etc.

Yeah definitely, that almost goes without saying for the most invisible NR.

[cinetj]

Hi Danijel,

I've tweaked a lot the "musical noise" and the "whitening" stuff in advanced mode that correspond to the smooth command... Strangely when I raise the musical noise there is a limit where the artifacts begin... My other bet was the treshold because of what I said that the reminiscence of the noise is dynamics related... Maybe there isn't much I can do anymore and it's better to leave a little noise...

Uncle Bob, I know small bits are better... But the othe benefits of Rx denoiser is that it erases tonal and broadband noise at the same time precisely my case here. I think I'm fighting a computer vent...

[Jfriah]

Are you using the plugin within a DAW or as a standalone plugin?

And always always always, you're better served by doing multiple passes with less noise reduction. If you're, say, taking off 12 dB in your initial pass, try reducing that to 8 and then do another pass.

Keep a sample of the 'train' section you used on a separate track for easy reference (or set up markers for the section).

ALSO---I find I get much better (though takes MUCH longer) using the C-offline algorithm. If you think about it, you're asking the software to make the changes in real-time if using Algorithm A. I find it better to let the software analyze everything more in-depth and process it if I have the time to walk away for a bit.

Additionally, using the spectral mode can work WONDERS because you can see in a nice waveform/spectrogram display and also LOOP JUST THAT FREQUENCY/NOISE. But it sounds like you are using the plugin within editing software (pro tools?).

And then there is the old argument: if you are making up gain anywhere, do you reduce the noise BEFORE or AFTER raising the gain of the file/section?

-Jeff

[cinetj]

Hi Jeff,

Yeah I'm using it at Protools. Although I have the standalone version also an already looked it up at the spectogram; the tonal stuff appears as a horizontal line around 500 HZ. Do you mean I can only use that for analysing noise at the standalone plugin?

2 passes and the other algorithm. Great tips, thanks. I'll try them.

Rodrigo

[ErikG]

If you can see the tonal issues a s a clear band around 500Hz with no major overtones Try to notch before using Noise Reduction. Notching and NR often compliment each other, not relying on one technique only often helps.

If its to wide to Notch, using spectral repair can do wonders.

After all the extra noise reduction is done, a little multiband expansion can often lower the low level artifacts enough for it all to work.

What I wouldnt give to get really pristine dialog recordings...

Here in Sweden the dialog editors prep and "almost premix" much further than what is the norm in Hollywood.
But only do so after you have experience in hearing how it all translates on a dubstage, and after talking to your mixer.
Undoing can be a mess during mixing.
Make sure that you have speakers that work for you. I find that on Genelecs (wich I hate) I would overwork artifacts and noise and thus have to bear this in mind whenever I do work on them.

[cinetj]

Hi everyone,

Thanks for the tip, Eric. In fact I've followed the sugestions above and did:notch filtering (RX Hum removal), 1 pass denoiser, second pass denoiser.Sending the files attached. It seems that I can not supress yet all noise ands I can hear a bit the artifacts (especialy some low end lost) but it sounds already much better than my previous tests.


I couldn't use algorythm C: the preview is impossible. B works and I've used it instead.

RX shows messages like: "noise profile is innacurate: converted 44100 Hz"; "the same and the original files is mono"; "the same and the original file is stereo". All in Audiosuite mode. Also I've noted comparing output noise (key listening) and bypass in a pure sample of noise that the noise registered by RX lacks some information. Should I worry?

[Jfriah]

Hello again;

You very likely will not be able to hear the preview using Algorithm C-- remember it is not real-time. It basically is the same as an A preview but ends up with better quality. Try it on a short section and compare with your own listening devices (ears) and see what you think--if you trust it is doing a better job, it is worth the time to hit 'process' and come back to it.

I will listen to your files as soon as I can as I'm always interested in hearing what others do with noisy tracks (the nature of our work lives!).

Yes, if you can see 'straight lines' in the spectographs, that is information that you should be able to notch out using a sharp Q- Eq setting with any other plugin, or, yes, Hum Removal provided that frequency exists in the range of the hum plugin.

I recently worked on a project where I completely erased such low-end hum information in the standalone spectral repair version and also completely removed birds tweeting throughout the dialogue track. Needless to say, the director and producer were very impressed. As was I! But it is slow going. Individual events, limited to 4 seconds at a time.

-Jeff

[Jfriah]

Originally Posted by ErikG

What I wouldnt give to get really pristine dialog recordings...

...you'd give up your pay because 'they' wouldn't need you!


But I hear ya. Is it me or are things getting faster, cheaper, rougher-sounding? (rhetorical question)

p.s. Sweden--I am a big fan of "Let the Right One In" and thought it was a great mix and they did some great stuff with the location sounds.

-Jeff

[Neil Whippey]

Yeah, I'm using RX at the moment for restoring Dials. they were recorded outside Liverpool Street station in London (a very busy area) and the background is constantly changing, so it's proving very difficult to get a clean dialogue without "musical" artifacts. luckily the atmos and spot effects that i've added have hidden the artifacts, but I'm still not particularly happy with it as I'm having to run these slightly higher than i would have wished and to my ear i can still tell just about that there's edits :/

ADR isn't possible either as both actors are quite well known here and did it for free because they liked the script - Oh for a Cedar right now!!! Is there anyone on here in London who'd let me run the clips through one????? Pretty please with a cherry on top......

[ErikG]

Originally Posted by Jfriah

...you'd give up your pay because 'they' wouldn't need you!


But I hear ya. Is it me or are things getting faster, cheaper, rougher-sounding? (rhetorical question)

p.s. Sweden--I am a big fan of "Let the Right One In" and thought it was a great mix and they did some great stuff with the location sounds.

-Jeff

"Let the right one in" was mixed partially at our place but not by us unfortunately... Would have been fun to work on it. I just finished mixing the score for a german movie with the composer that wrote the score for LTROI.

[matt-o-]

I don't know if it is common, but I get very good results by not using the denoising on the full spectrum in RX.
I prefer to use the standalone application, although it makes the workflow a bit more complicated: in PT consolidate (shift-alt-3) problematic clip - open file in RX.
With the vertical ruler select noise sample, train, then switch to horizontal ruler and select only the middle part of the spectrum, from around 100 to 300Hz til 6 to 8kHz, and apply. That keeps the deep parts of the spectrum from "warbling" and the untreated "highs" stay more natural and open. Then I'll save and voila - as PT still links to the consolidated clip, the treatment is immediately reflected in the clip in the timeline (but maybe not the waveform). This way I avoid double IDs. I do not denoise too much, between 6 and 10dB maximum.

The lows are then hi-passed as necessary in PT and eq'd.

Greetings,
Matthias

[cinetj]

Sorry Eric, I've passed right over your other commentaries. I don't have excellent conditions. I work in my home studio with a pair of Adam A7. It's no Genelec (that you already dislike) but I left the original sound backed up in a track within the session. That way if I screw things up we can still turn back...

Jeff, I really appreciate if you can listen and give me your opinion. Just remember: audio poorly recorded, not edited, not nothing (only denoised, I hope). So take it easy please... I'll try the C as you told. Sure it won't work in the preview... how stupid am I? Don't answer that one.

Seem a lot o work Mathias. But if it pays... Izotope could have made everyone's life easier if the spectral analysis was included in the plug in version...

[Jfriah]

Hi; I listened to the clips today and worked on some of my own but got tied up with another project. Will post results back on Thursday.

I found the same thing, though, the noise rode along the voice quite heavily and even with gating, etc. I still got noise remainders at the end of lines; and the male voice was quite soft spoken at the end of phrases.

I think I had the best luck (and completely erased the buzz and some mouth noises and the voice of the woman at front of clilp) with standalone application of iZotope RX in spectral mode, then following that with a couple passes of Waves X-Noise. iZo's noise reduction seemed to be too warbly too easily.

Anyway, more on Thursday.
p.s. I also ran it through a rough CEDAR pass and it helped quite a bit but...

-Jeff

[cinetj]

Hi Jeff,

Thanks for the effort!

Yeah the end of the lines is critical regarding the broadband noise. One thing I didnt do and that was suggested here by Eric was the multiband expander. That should get rid of the rest of the noise. Waves still worths a shot then? Good to know.

By the way, I was concentrating in the broadbandnoise tonal stuff. I didn't mean the other problems (mouth clicks etc). Also the file I sent is trimmed way farther what I'm gonna use it. The voice of the girl I'll just edit it for example.

Talk about obssessive...

[Airon]

RX is often part of the solution for me. It's tonal noise removal is a godsend. Nice tip on the limited bandwidth. I'd been treating the the RX a little like the Cedar since they put envelopes in to the denoiser. Hadn't thought of that one though.

Btw, on an i7 920, I could live-preview the C algo in Protools 7.3.1 in the Audiosuite version of the denoiser.

[cinetj]

With a processor like the i7 it's possible then. Unfortunately I don't have that much power...

[Matti]

My version made in couple of minutes. RX, 1 pass, could do getter with some tweaking

Matti

[cinetj]

Hi Matt,

Impressive;It's way much better than mine. What did you tweak exactly? No other processing really?

[Matti]

Nothing else, I didn´t save my settings but I was in advanced mode.
I´ll try to recreate it and show you

Matti

Edit. Here, it was something like this -notice the FFT size for this kind of noise. You can also see what I used for training the thing.
-quite hard settings eh?

[Jfriah]

That's ONE pass of Rx? Wow, nice job! I was just about to post 5 files with various steps including Rx, Waves X-Noise, and Cedar but I won't bother now, hahah.

What I found on everything was the tail of the voice still had the noise riding from the tape machine or whatever it was recorded on. Still a bit there / bit of artifacting but it is night and day compared to the original file! Nice work!

(going to go try with your posted pictures)
-Jeff

p.s. were you in standalone mode?

P.P.S---haha, you forgot to mention by ADVANCED MODE you meant the $800 more expensive ADVANCED VERSION!

[Matti]

I´m sure we could make it better with some time to adjust, that was done in some minutes for that file -because i couldn´t believe the other samples

Matti

[Jfriah]

Originally Posted by cinetj

Hi Jeff,

Thanks for the effort!

Yeah the end of the lines is critical regarding the broadband noise. One thing I didnt do and that was suggested here by Eric was the multiband expander. That should get rid of the rest of the noise. Waves still worths a shot then? Good to know.

By the way, I was concentrating in the broadbandnoise tonal stuff. I didn't mean the other problems (mouth clicks etc). Also the file I sent is trimmed way farther what I'm gonna use it. The voice of the girl I'll just edit it for example.

Talk about obssessive...

No worries---I like to play with other peoples' files when I get a chance; keeps me sharp and practised when I'm between fixing my own stuff (!!!). Also gives a great opportunity for everyone to share tips and tricks which, I think, only serves the greater good.

I was doing the mouth clicks and errant noises just to show what else can be done with Rx--like I said I previously had removed bird chirps from a scene in a feature, completely. And some brake squeaks.

Nice plugin...now have to recycle some bottles and cans and save up for the Advanced version!!!

-Jeff

[Jfriah]

Originally Posted by MATTI

I´m sure we could make it better with some time to adjust, that was done in some minutes for that file -because i couldn´t believe the other samples

Matti

So that was Rx only --- (again, was it standalone or plugin mode?) and without a multiband expander (which I'm still always trying to get the hang of---did some stuff with a C4 a few days ago but it was too-affecting)

-Jeff

[Matti]

Nothing else used. Check the "normal" RX, I think it has also some advanced options. Standalone this time.

Matti

[matt-o-]

OK, here are 3 other offers:

one RX treated, based on the setting from MATTI, but with the normal, non-advanced, version. As described before, one pic shows the spectrum, where RX worked: from 230 - 5500 Hz, the extremes stayed untreated, the other shows the settings.

Then, for fun, two versions with the Dolby 430. As I really had to pull the "high" fader down to the bottom, I cascaded two units for the other take, with both high faders pretty down, but low faders stayed nearly untreated.

JFYI,

Greetings,
Matthias

[Matti]

The second Dolby sounded ok aswell!

Matti

[cinetj]

Definetly, Matt's trick is the winner! Thanks, Matt!

Narrowing the area of the plug in's action sure makes a difference. And it's so good to know now, thanks

Just for curiosity I send one last sample with the same parameters that Matt used but without the band action limit. You can hear the artifacts clearly already in the first sentence.

Jeff, regarding the multiband expander, it isn't anything promotional but I'm fine using Izotope (yeah, once again) Ozone's multiband dynamics.