Learn Music and Audio Production | iZotope Tips and Tutorials

5 Tips for Noise Removal with RX 8 Elements

by Mike Metlay, iZotope Contributor February 17, 2021
Try these challenges to sharpen your skills

Equip yourself with essential audio repair tools:

RX Elements

iZotope email subscribe

Never Miss an Article!

Sign up for our newsletter and get tutorials and tips delivered to your inbox. 

It’s time to turn our attention to RX Elements and learn how it provides easy and effective noise removal. In these tips, we’ll take a stroll through its four plug-ins, and wrap up with a look at just one of the other useful tools that come along with the package. Let’s get started!

1. Make the most of De-click

First up is the De-click module, which provides an easy way to go after one of the more common artifacts in digital recording: the infamous click. When you ask your DAW to do something it can’t—such as record audio data faster than your interface can pass it along, or pull just a hair too much power from your CPU at a critical moment—the flow of audio to your storage medium gets interrupted, and you get a click. 

Clicks can come from all sorts of sources, but the ones you’re most likely to run into are those that come from things like buffer underruns and processor spikes. 

Here’s the interface for the De-click module:

The De-click module is easy to set up for detailed de-clicking: the critical parameter here is Sensitivity.
The De-click module is easy to set up for detailed de-clicking: the critical parameter here is Sensitivity.

You can choose single-band or multiband operation. The multiband Frequency Skew gives you finer control when hunting down low or high frequencies in cicks. Click Widening lets you smooth out clicks by pulling data from the area on either side of the click; how much you use is up to you and depends on the nature of the program material. (I usually keep it set to 1–2 milliseconds.)

The critical parameter is Sensitivity. Set it all the way down at 0.5, the module will barely react to any clicks at all; all the way up at 10.0, it will interpret nearly any glitch as a click (dozens or even hundreds per second!) and try to fix them all, which can do some pretty weird things to audio that’s not full of actual clicks. (I usually start between 1.0 and 2.0.)

In the example below, we hear a snippet of audio from a planetarium soundtrack project I worked on a while ago. Note the odd waveform present in much of the snippet—I deliberately overloaded the input stage from one of my instruments to provide a bit of soft clipping. However, the clipping got out of hand, leading to a nasty click—seen as a vertical line in the frequency spectrum. Since it’s broadband, I used the Single-band algorithm and played with the Sensitivity and Click widening to get rid of nearly all of the click while preserving the surrounding audio.

Here are before-and-after screenshots of the audio, and what they sound like.

The click in the audio is visible as a vertical line in the spectrogram, meaning it has all frequencies represented.
The click in the audio is visible as a vertical line in the spectrogram, meaning it has all frequencies represented.
The De-click module (with the settings shown above) pretty much erases the click, leaving only a tiny thump.
The De-click module (with the settings shown above) pretty much erases the click, leaving only a tiny thump.

Before De-Click

After De-Click

Pretty dramatic, isn’t it? I could fine-tune it further, but this example shows how just a couple of minutes’ work can produce very usable results.

2. The miracle of De-clip

The De-clip module is taken for granted by hardcore RX users these days, but when the function was first introduced several versions ago, it seemed like black magic!

It was previously believed that once waveforms suffered severe clipping, often from hitting an analog-to-digital converter with too loud a signal, they could never be recovered with anything like the accuracy of the original unclipped waveform. De-clip takes advantage of the enormous power of even the most inexpensive modern computers to make this formerly-esoteric process not only feasible but easy.

Here’s the user interface for the De-clip module, shown at four points in our de-clipping process for the example below. 

 From left to right: an unclipped waveform; the clipped waveform in the example; the suggested De-clip threshold; and the resulting waveform.
From left to right: an unclipped waveform; the clipped waveform in the example; the suggested De-clip threshold; and the resulting waveform.

The Threshold diagram features a histogram (a fancy word for a distribution of values of a given parameter) for waveform elements in the piece of audio we’re examining. In the first picture, we see a conventional unclipped waveform. The histogram shows that nearly all of our audio samples are well below digital zero (0 dBFS, the maximum level an analog-to-digital converter can handle before it clips).

The second histogram tells a very different story, borne out by a view of the audio itself:

Note the places where the highest peaks have been ‘sawn off’? That’s clipping, and it sounds as bad as it looks.
Note the places where the highest peaks have been ‘sawn off’? That’s clipping, and it sounds as bad as it looks.

This is a very dynamic passage of solo digital piano, played during a live set several years ago. I am leaving the performer anonymous because of the poor quality of the recording. That’s not his fault but mine: engineering from the stage under the pressure of a live performance, I mis-set our audio interface and the piano performance was clipped at –7 dBFS going into our recorder, resulting in some loud peaks being clipped off. Not too many years ago, the audio would be considered a complete loss. Now? Maybe not so much.

The second UI image above shows what an analysis of this audio provides: brutal clipping, with way too many samples in the affected area crammed up against -7 dBFS. In the third image, we see the suggestion De-clip has made: to analyze and repair anything it finds above -7.1 dBFS.

The Makeup gain slider lets us reduce the final level of the audio, as the repaired peaks may (and often will) be far above the dynamic range of the existing audio. It’s set pretty low here, because the clipping occurred at such a low level, it’s unlikely the repaired waveforms will still clip within the actual dynamic range available to this audio file.

Here’s how RX Elements displays what the plug-in is about to do.

Once we choose a De-clip threshold, the main display shows how that setting relates to the clipped audio.
Once we choose a De-clip threshold, the main display shows how that setting relates to the clipped audio.

Using the suggested settings resulted in a largely-repaired performance–-not perfect but surprisingly good, and this was a pretty nasty example to start with. The fourth UI screenshot (above) shows the new histogram, and here’s what the resulting audio looks like, before and after we close the De-clip window:

The boundaries of De-clip now show where the repaired peaks are pretty clearly...
The boundaries of De-clip now show where the repaired peaks are pretty clearly...
… and when we close the De-clip window, the graphics go away, showing our repaired waveform.
… and when we close the De-clip window, the graphics go away, showing our repaired waveform.

Here are the before-and-after audio clips. It’s a pretty dramatic difference!

Clipped Piano

Piano after De-clip

De-clip is surprisingly effective on passages like this, where clipping doesn’t occur everywhere. A ‘bricked’ recording with everything slammed up against 0 dBFS will be a much tougher proposition, and re-recording the music will almost certainly be your best option. For when you can’t, though, De-clip can at least have some effect when restoring your audio.

3. De-hum does what it says on the label

The trick to using the De-hum module is to realize that it’s tightly focused on cleaning removing one and exactly one kind of audio nastiness: hum, a (usually) neatly-ordered series of harmonics that comes from the presence of an unwanted source of interference in a studio.

A good rule for laying out cabling in a studio is: “power crawls, data walks, audio flies.” In other words, where it’s feasible, keep your power cables and wall warts on the floor or as close to it as possible (running power cables from gear straight to the floor); run USB and MIDI cables a couple of feet above them if you can (and down on the floor with them if you can’t); and run audio along your racks and keyboard stands at the level of your gear, keeping it away from data cables and really far away from power cables and supplies. When power and audio cables must cross, do it at right angles to minimize the pickup of hum in the audio. Ditto data cables to minimize digital noise.

However, even these precautions can still leave you with various sorts of hum in your audio, and most small studios and home recording setups won’t be able to all of this, or even any of it, if power and data and audio all have to run into the rear panel of a small interface on a desktop, with cables in parallel and quite close to one another. 

Fortunately, De-hum does a bangup job of removing hum as long as you recognize its limitations and use other means to supplement it. This is what the De-hum interface looks like:

The De-hum module has a series of notch filters to remove hum harmonics precisely without harming audio.
The De-hum module has a series of notch filters to remove hum harmonics precisely without harming audio.

Clicking the Learn button lets the module find the base frequency and harmonics, setting its surgically-tight filters to attack the relevant frequencies. You can choose up to 16 harmonics to treat, and even turn certain harmonics on and off depending on the nature of the hum (less processing is always better in cases like this). You can also improve accuracy sometimes by linking only the even or odd harmonics rather than all of them.

Note the displayed waveform under the filters. It shows where the analysis has found the worst hum and what the final waveform will still contain after processing. It’s pinpointed 60 Hz as the fundamental, and it does show some remaining very high harmonics–-that’s why the plug-in is prompting us to perhaps try a different approach. However, as we’ll hear, De-hum produces great results anyway!

Here’s the audio we’re working with, and no, there’s no music in there. It’s a sample of pure hum, generated by the worst possible scenario I could create in my studio: I put my Schecter A-5X guitar on the floor near the carefully isolated rat’s nest of power cables and supplies in the far corner of the room, ran its guitar cable over and through the power cables, and recorded what came out at very high gain. I had to do this after hunting through my raw recordings for several hours trying to find an actual music sample with hum in it–-I was forced to accept the fact that following my own rules seems to have had the right effect!

The harmonics of pure power-line hum. Pretty disgusting, right?
The harmonics of pure power-line hum. Pretty disgusting, right?

I then set up De-hum and ran it, with the settings shown in the UI screenshot above. What I then learned, via turning various harmonics on and off, is that this particular sample of hum was strongest in the odd harmonics, with the 3rd and 7th being particularly powerful.

To illustrate the effect of De-hum, I went through the sample and processed different sections of it, gradually adding harmonics. In the image below and in the second audio sample, you will see and hear what I’ve done. The raw hum alternates with De-hum attacking 1, 2, 4, 6, 8, 10, 14, and finally all 16 bands. 

You’ll notice that removing 1 vs. 2 isn’t very different at all, due to the weakness of the 2nd harmonics. However, 4 is a big improvement (having taken out the 3rd harmonic), the difference between 6 and 8 is dramatic (having taken out the 7th), and with a full 16 harmonics dealt with, the only thing left is a bit of high-frequency buzz that will be well below the level of any musical audio in the track.

The spectrogram shows the removal of 1, 2, 4, 6, 8, 10, 14, and finally all 16 hum harmonics.
The spectrogram shows the removal of 1, 2, 4, 6, 8, 10, 14, and finally all 16 hum harmonics.

Here are the samples to review:

Before De-Hum

After De-Hum

Maybe something like Spectral De-noise or Guitar De-noise (available in the full versions of RX 8) would do a better job here, but on its own, De-hum is a powerful tool for audio repair as long as you understand what it can and can’t do.

4. Voice De-noise, for our most important sound source

We can all agree that for practically any studio work these days, getting the best quality in our vocal tracks is paramount. The human voice is the first thing we learn to listen for as babies, and it leaps out at us as something we instinctively need to understand. Our ears are actually tuned to the frequencies of human voices—check out the Equal Loudness Contours to learn more about the “built-in EQ” that our ears have. As a result, when something’s wrong in a vocal, even an untrained listener will notice it immediately.

With the explosion of social media streaming and other online audio, getting the cleanest possible vocals is something everyone needs. RX Elements provides a comprehensive solution in the form of the Voice De-noise module.

As you can see from the screenshot below, Voice De-noise puts a lot of power into a very simple interface:

Voice De-noise combines equalization with noise reduction for optimized effect.
Voice De-noise combines equalization with noise reduction for optimized effect.

After selecting whether you’re working with a tonal voice (a vocal in a song) or an atonal one (spoken word), you can hit Learn and let the module analyze the voice. Sometimes you’ll want Surgical filtering to home in on specific frequencies at issue, and sometimes you’ll want Gentle filtering for a more broad effect. Then just set the Threshold where processing kicks in, crank up how much noise reduction you want without creating artifacts, and Render the result. It sounds complicated, but with only two buttons and two sliders, it shouldn’t take more than a few tries to preview and then apply the best results.

I recorded some sample dialogue with HVAC noise in the background, then hit it as hard as I could with Voice De-noise. The results are pretty darn impressive, the HVAC is effectively gone, with little or no impact on the voice quality.

Before and after Voice De-noise, visually and audibly:

Before Voice-Denoise, with HVAC background noise visible between words...
Before Voice-Denoise, with HVAC background noise visible between words...
… and after Voice-Denoise, the background noise is effectively gone and the vocal quality is barely impacted.
… and after Voice-Denoise, the background noise is effectively gone and the vocal quality is barely impacted.

Voice before De-Noise

Voice after De-Noise

Voice De-noise is an incredibly powerful noise removal tool, and perhaps offers the best bang-for-the-buck of all the modules in RX Elements. Because it is optimized for vocals, that’s the source it’s recommended for, so I would never dream of suggesting that you might try it out on other midrange-heavy material like acoustic guitar or wind instruments, to see if it works for them in certain cases. Sorry, can’t do it.

5. Mixing–just one of the other toys in the box

When I got to the last tip in this article, I was faced with a bit of a dilemma. RX Elements has four main modules, but I wanted to provide five tips. What to do?

That’s when I got to thinking about what else comes with RX Elements in its standalone application. In addition to the four modules described above, RX Elements offers much more in the way of utility functions—everything from fades and normalization/gain adjustment to phase rotation and the ability to host third-party plug-ins. I should also mention Repair Assistant, a one-click way to get suggested starter settings for any given cleanup job.

Just giving a laundry list is fine and dandy, but wouldn’t it be nice to wrap up with a really creative use of one of these modules, despite their relative simplicity? It would!

The Mixing module, set for (left to right) conventional stereo, summed mono, and phase-inverted stereo.
The Mixing module, set for (left to right) conventional stereo, summed mono, and phase-inverted stereo.

This is the Mixing module. On the surface, it looks incredibly simple: you have the ability to set how much of the left and right input audio you hear in the left and right output signal, and then render the result. But therein lies the fun part: you can control each side not only in its own output, but in the other one–-and you have the option of inverting signals as well. That opens up a lot of possibilities!

In these UI screenshots, we see just three of the practical settings for Mixing. On the left is the conventional stereo setting: all the left in the left, all the right in the right, nothing from the opposite channels. In the center, we feed all of both channels to both outputs: in other words, we’re summing to mono. (Some engineers say that the best way to hear mono is to feed one or the other side to one or the other speaker—this is easily done here as well, and you can even save these various settings as module presets.) On the right, we are subtracting the cross-fed channels, inverting them with respect to the primary channels, producing an output of audio that is out of phase between left and right.

This can produce all sorts of interesting audio effects if dialed into various degrees: you can narrow a stereo field a little (or a lot), or bring out phase-inverted elements for more perceived stereo. There are also diagnostic tricks for learning about your audio chain, as in this example.

My studio has a small but high-quality interface for simple tasks (like creating these articles). It’s the Yamaha AG06, a 6-channel analog mixer and stereo 24/192 audio interface. The first two channels have built-in effects, including a surprisingly good reverb. These effects can be programmed from a computer control panel and saved in several user program slots, one of which can be offloaded and stored in the mixer for when the control panel isn’t running.

The reverb in the AG06 has a Diffusion control. Every reverb algorithm has a slightly different way to implement diffusion, and I have always wondered what this one did. With the Mixing module, I was able to find out in seconds!

The first three screenshots and audio samples are with the reverb’s Diffusion set to 0 (turned off). In the original stereo mix, it sounds dense but centered, making me think it’s a mono reverb applied to the mono mic signal. Using the second set of Mixing settings, we hear the same signal but 3 dB louder, a perfect example of summing two identical channels. And with the third set, the entire signal, both dry and wet, effectively vanishes, a Diffusion of 0 is effectively mono.

Here’s our signal with the reverb Diffusion set to 0, as a standard stereo signal.
Here’s our signal with the reverb Diffusion set to 0, as a standard stereo signal.
Here’s the same signal mixed to mono. It’s unchanged, just 3 dB louder, as you’d expect from summing identical signals.
Here’s the same signal mixed to mono. It’s unchanged, just 3 dB louder, as you’d expect from summing identical signals.
Here’s the same signal with full phase inversion. Poof, gone!
Here’s the same signal with full phase inversion. Poof, gone!

Reverb 0 Diffusion Stereo

Reverb 0 Diffusion Mono

Reverb 0 Diffusion Inverted

In our final set of examples, we crank Diffusion all the way up. Now we have a very different stereo sound, more spread out than the previous example. What does this mean in terms of the actual signal processing?

When we sum to mono, the dry vocal is 3 dB louder, but the stereo reverb has a very different character: it almost sounds like the mono version with Diffusion of 0! What’s going on here? The answer lies in the phase-inverted version: the mono vocal effectively vanishes, and the reverb remains. That implies a huge (if not 100%) proportion of phase-inverted audio to begin with… when we reverse that reversal, we cancel the dry vocal but intensify the stereo reverb. Pretty cool!

Here’s our signal with the reverb Diffusion set to 0, as a standard stereo signal.
Here’s our signal with the reverb Diffusion set to 0, as a standard stereo signal.
Here it is in mono...
Here it is in mono...
...and here it is, phase inverted. The dry vocal is gone, but the reverb remains.
...and here it is, phase inverted. The dry vocal is gone, but the reverb remains.

The takeaways

With four useful modules that address the most common noise problems in small studios, RX Elements offers a ton of options on a very small budget. With these great approaches to noise removal and a host of other useful utilities, it’s pretty much a no-brainer for any small recording rig. Check it out for yourself!

iZotope Logo
iZotope Logo

We make innovative audio products that inspire and enable people to be creative.