Join Scott Hirsch for an in-depth discussion in this video Using Denoise for broadband noise issues, part of iZotope RX 4: Repairing and Enhancing Audio.
- In the movie, you were moving hum and line noise, we saw briefly how the Denoise module can help hum issues. In this movie, we'll use it for something that it also excels at, broadband noise, like traffic wash and air conditioning fan noise. This audio in the first example has broadband noise. That is, the noise is scattered all over the frequency spectrum. Let's take a listen. - Fasten your seat belts and prepare yourself for atmospheric reentrance. - We hear a prominent air conditioner noise, in addition to the voiceover.
Let's open the Denoise module. Now the key to making the Denoise module work, is to find a section where there's just noise. The module can then set up a noise profile to use, in order to extract the noise from the usable audio. Luckily here in this example, we have a space before and after the subject speaks that we can use to set up a noise profile. So we want to make a time base selection across that spot where it's just the noise and then we click learn in the Denoise noise module.
You can see a noise profile in orange is set up. Now, you don't always have the luxury of a contiguous space where there is none of the audio you want to keep. If that's the case, there might some areas where you could use the time frequency selection tool. So in other words, I could use this to select just the high end, the high frequency stuff here, if I had an area where I could grab it from there. Then I can hold shift and I can go elsewhere and maybe grab the mids from somewhere else, and I can go elsewhere and maybe grab just the lows from somewhere else.
So, if you have a piece of audio where you're able to grab different areas from different spots, that will work as well. You can go ahead and learn this, and in this case we do have some missing spots. You can see there's a gap there, there's a little gap there, but you'll notice the Denoise module asks us noise profiles not captured at some frequencies, complete it? And we'll say yes, and it actually fills in the blanks and it completes the areas that we missed in our selection. So you might not have everything, but you wanna get as much as you can even by using this method where you're selecting, holding shift to select different areas.
You can learn noise profile that way. So that's a pretty cool thing you can sort of interpolate in noise profile. But, let's go back and I'll use the time selection tool and grab the noise profile and I'll click learn. Once we have a good noise profile, there's some other settings. First of all, this graph here shows us visually what's going on with the noise profile and the residual noise. When I click preview, - pheric reentrance. - I can see some other information, namely the input and the output.
So that tells us what's coming in, in gray, and what's going out of the module in white. Then you can see where the residual noise lies in terms of the output. Now this graph is pretty useful in general. I guess visually, but really it's gonna take your ears to know if the noise reduction is actually working, or not. Speaking of listening, really the name of the game in noise reduction using the Denoise module is noise reduction versus artifacting.
It's usually possible to reduce a lot of the noise in any given audio clip, but unfortunately, that can come with the added expense of having the original audio sound artifacted. It'll sound glitchy, underwater, chirpy, these are all sonic artifacts that will be introduced if you go too far with the noise reduction. So how does it work? I usually start first, by upping the quality slider to best. Again, as a trade off with time spent rendering, and best sound quality.
Usually, I'd rather sacrifice time for better sounding audio. So I usually go for the best. Next, I'm gonna make a selection across the whole thing, now that I've go the noise profile set up. You want to deal with these parameters over here. Namely the noise reduction. So that's gonna be the amount of reduction. If you go higher, you're spreading the gap between your noise profile and you're pushing the residual noise down. So higher reduction values mean more reduction, potentially more artifacting. And lower reduction values, which is the residual noise closer to the noise profile, all the way down, is no noise reduction.
So, if you option, or alt click, for Windows on the slider, you get a reduction value of 12. That's the default setting. In most case scenarios, I'll have to say, usually I'm going lower than 12, because it's too much artifacting at 12. Usually just a little bit less, somewhere around nine or eight. In this example, I think because our noise profile was so clean, and it doesn't change over the whole course of the example, we can actually go higher than 12. Which, if you can go higher than 12, you're sitting pretty, because you're really gonna get some nice noise reduction.
Just with that basic setting there of 18, we can take a listen. We can preview it and see if it's working. - Fasten your seat belts and prepare yourself for atmospheric reentrance. - Now, that's I'd say, pretty dramatic. I mean we didn't hear any of the air conditioning going on. But, you know, another way to deal with this as you're previewing, and one thing that I definitely recommend is you want to choose output noise only. That way you're hearing the difference. You're hearing only the noise, right? Which is something that you want to do because when you're previewing, you can listen to the noise, and hear only the noise.
If you start hearing, if you're moving the reduction slider up and down while you're previewing, and you start hearing the voice come in or the source material that you want to keep, maybe it's a guitar, maybe it's a bass guitar, or whatever you're working on, you start to hear that creeping in, when you're outputting noise only, that means you're removing some of the source material. Which, you don't want to do, and there's a higher potential that you're gonna get the artifacting. So, that's something that I would definitely recommend, is output noise only, listen to it while you're moving the reduction slider up and down, and find that sweet spot, where you're getting the most reduction you can, without getting into hearing any of that artifact.
Couple other nuances here, you can click up here where it says Curve, this blue dot, when you click that, it actually gives you a way to weight your noise reduction based on frequency. So, if you want to deal with more noise reduction over high frequency, say from 5k, up to the top there, you can pull this down. It's gonna actually make more noise reduction happen. More denoising happen in the higher frequencies. Or, the opposite. If you want less noise reduction, you can put this above.
You can do this anywhere. You can just pick an area in the middle, for example. You can really customize the curve and sort of weight things differently, depending on how you want to work. Depending on if there's a lot more high frequency stuff you need to get rid of, for example. Or, you can uncheck this and it goes back to just equal weighting across the frequency spectrum. Two other parameters threshold, I usually generally leave threshold alone until I get the reduction going the way I want to, and then I can maybe eek out a little bit.
If I go down, it's gonna make more noise reduction happen. It's lowering the threshold, and the noise profile. If you go above zero, it's actually making less noise reduction happen. But generally, in my workflow, I keep that at zero and do the reduction first. Then I'll maybe eek out a little bit more noise reduction by moving the threshold slider. Then finally, when you start getting artifacting, if you just cant get it enough, it's almost perfect, but there's still a little bit of chirping, a little bit of that artifacting, we have this parameter here.
It's called artifact control. In this one, basically, to the left, we'll sacrifice, it'll make better noise reduction at the expense of hearing more chirping. So, if you need more noise reduction, you can try to move this to the left, but the result might be that you get some of that artifact. And likewise, if you move it above the default setting of seven to the other side, you may potentially get less of that chirping, but you might get more of the noise that you're trying to reduce.
So, in general, I think seven's a good place to start. But you can sort of move this up and down. Again, it's all fine tuning nuances that you really have to preview. It's going to be different for every type of audio that you're working on. These are things that you want to do in preview mode, while you're listening to the noise, and then take that check off and listen to the actual source audio, and just try to get it, that sweet spot where you're getting the best signal to noise ratio. You're getting more of the original sound with less of the noise and as little artifacting as possible.
Now two other things that I want to look at here, we've been in manual mode, which means we manually set up our noise print and learn it. There is an adaptive mode. So, if I choose that, basically there's no more noise print. Basically, the module looks at incoming audio and sets up the noise profile based on that. It has learning times. You can make it learn faster or learn over a longer period of time. The adaptive mode, in my experience, is really good for any type of audio where the background noise is changing over time.
So, for example, if you have a longer selection than the one we have here, and maybe there's traffic that's changing over time, one single noise print, in manual mode, isn't gonna cut it, you can choose adaptive mode and make the module learn, in real time, on the fly. That can be very useful for that. The other thing that I wanted to look at, is the advance setting. So, here's some very advanced nuance parameters for someone who really wants to get in under the hood, you can change some of these parameters to really try to fine tune the Denoising artifact and try to really get in there and fine tune.
But one thing that I wanted to show you here, is under quality, as I start moving from A, B, C, and D, from fast to best, notice that these parameters are actually changing. So that's what's happening in detail as you move the quality from fast to best, is actually some of these advanced parameters are being changed. Now, I want to use, notice that we've been using the Spectral tab here, and that's good for broadband noise, in general. There's also one that's specific to dialogue. If I click over here, we get the dialogue tab.
Then we just move to the default setting here. I'm gonna use the dialogue tab of the Denoise module on the second example, which is broadband New York City dialogue. It's a documentary. It's probably a camera mic. Let's take a listen. - A place where there's something there, and then I can sorta stay put, and let people walk through and they create the-- - So these are the types of clips where we can really do good work with RX. We'll allow more intelligibility and higher levels of the interview subject over the noise floor.
There's no way we're ever gonna get rid of all this noise floor, but the name of the game here is we want to just expand, basically increase, the signal to noise ratio. We want more of the voice, and if we have a way to push the noise down a little further to increase that space between the level of the voice, and the noise, then we're doing some good work. So, we're gonna use the dialogue tab. Now the dialogue tab, basically is the new dialogue optimized noise reduction in RX 4. The idea behind this tab is that it is a simple, no latency denoiser.
Ideal for achieving basic, high quality denoising on a variety of material with a minimum amount of time spent on tweaking controls. You'll really see the power of this denoiser, this dialogue denoiser module, when we work in a digital audio workstation, like Pro Tools. That's where the no latency really shines, as you'll see. But, when you're in stand alone RX, it's also great at anything dialogue related, like this clip here. It's a lot easier and simpler than a spectral denoiser.
You really have two options. You've got manual mode, and you've got auto mode. Auto mode's awesome 'cause it just basically listens to incoming audio, and it automatically optimizes the noise envelope based on it. Let's preview this and see how it works. - A place where there's something there, and then I can sorta stay put, and let people walk through, and they-- - So I can hear it's definitely reducing the noise quite a bit. I hear some artifacting, actually. So what I want to do, in this case, is just back off the reduction a little bit.
And, let's preview that. - A place where there's something there. - And I'll hit bypass-- - Then I can sort of stay put-- - Hear the noise come back in. - People walk through and they create the photographs. - So yeah, it's doing like I said. It's just pushing that noise floor down, and the voice still sounds nice and strong. So, it's just increasing the gap between signal to the noise, which is great. The other mode is Manual mode. That works just like Spectral, where in Manual mode, we're going to go in and find a noise print. We don't have much to work with. Maybe just this little section here, and we can learn that, and it sets up a noise profile based on that section.
But again, since this is New York City, and we have background stuff, it might be changing over time, I like the Auto mode, a little better there. It's able to deal with any nuance and changing audio over the course of the interview. Once we have that set up, go ahead and process that, and we've done the denoising there. Now, if you're a dialogue editor, or you're working in post production in any capacity, these two broadband denoisers, Spectral and Dialogue, they're heralded as the best in the business.
They compete right up there with much more expensive noise reduction software and hardware.
- Exploring the workspace and workflow
- Processing files
- Cleaning audio
- Removing noise
- Working with a DAW
- Cleaning up noise in music and field audio
- Utilizing EQ Match and Ambience Match to fill gaps and match audio
- Time-stretching and pitching audio