Izotope RX Advanced

reviewed by David Mondrup

Izotope presents RX as a complete standalone application designed for audio restoration. It comes in two flavours: RX (cheaper) and RX Advanced (much more expensive). Both of these can be bought through the Izotope website, and RX is also available in shops in a boxed version. The software can be authorized through a standard challenge-response procedure, as well as through the Pace iLok system. The copy kindly made available for this review by Izotope is the RX Advanced.
Since the beginning of the process of reviewing RX, Izotope made plugin versions of the 5 modules (mentioned below) available to registered users as a free download, thus enhancing the functionality of the purchase.

Audio Restoration.
As the presentation says, RX is made to restore audio suffering from one or more of a wellknown variety of audio illnesses. It comes with 5 different modules to deal with different types of situations:

• the Hum Removal module, a harmonic filter to deal with ground hum and likewise artefacts
• the Declipper module, to restore clipped audio
• the Declicker module, to remove clicks and crackles
• the Denoiser module, a noise profiling tool to deal with unwanted background noise and hiss

so far, no news to me. But then, there's:

• the Spectral Repair module, which rebuilds the audio in gaps caused by audio dropouts

That one is new to me – but then, I never claimed to know everything ...

Still, as suggested, the processes included here can hardly be called groundbreaking. With the exception of the Spectral Repair, I already had all the processes available to me as an Adobe Audition user. My best guess is, that any other wave editor on the market will also sport most, or all, of these functions. So, Izotope's selling point will have to be that they do these things better than others.

RX Advanced – the music box challenge
To test things, I decided to tailor a challenge for RX. An audio file recorded under so completely hopeless conditions that heavy restoration would be needed to bring it to the point where anyone would ever consider using it as a track in any kind of production.

I had just purchased a DIY Music Box out of thinkgeek.com. Having punched my first tune for the Music Box, I set out to record it. I used an old, broken PC microphone, which I had picked up with my first computer back in 1995. Back in the day, it set me back as much as $5, which was about 3 times it's actual value. The microphone connected through an unshielded, unbalanced minijack input to the onboard soundcard on my current laptop PC. This soundcard, carrying the Realtek brand, is no better or worse than any other onboard soundcard in standard PCs, which means that in terms of audio quality and build of A/D converters, it falls in a category somewhere between ”awful” and ”downright disastrous”. I placed the Music Box on the table next to the humming laptop, and held it down unsteadily with my fingers while turning the handle, listening to several loose items on the table rattling along with the actual tones transmitted from the Music Box through the table plate. Needing somewhere to place the ”microphone”, I laid it across the open laptop computer, right above the CD drive, resulting in magnetic charges inducing 2 highpitched notes alternating throughout the recording – a nice surprise that wasn't revealed until listening to the audio after the recording had taken place.
In short, I had everything set up for a technically absolutely horrible recording session. The result turned out like this:

music_box_1.mp3

Plenty of work for an audio restoration tool.

GUI
Before starting out, a bit on the graphic representation of audio in RX. Audio can be seen as the traditional waveform display, as well as a highly customizable spectrogram. The new thing, at least to me, is that the waveform and the spectrogram can be viewed at the same time, on top of one another. An opacity slider crossfades between the two views, from all vaweform to all spectrogram. Each view can be zoomed vertically independantly of eachother. Being able to watch both overall volume information as well as frequency representation at the same time has proven to be very useful, and also very easy to get used to.
As for selection tools, one can select horisontally (along the timeline) as well as vertically in the spectrogram view, singling out a frequency band. It is also possible to combine the two modes of selection. A special playback button allows for playback of only the selected frequency band. There is no way yet to apply freeform selection or have multiple selections simultaneously, which could otherwise prove useful for selecting audio for profiling in the denoiser module. Perhaps future upgrades will cater for this need?

Now, onwards to making that music box sound better:

The 5 modules
First off, I'd like to get rid of some of the buzzing, electric humming going on in this recording. What better tool for the job than the Hum Removal module? This works as a harmonic filter, doing a dip of a narrow frequency band and its harmonics. All sorts of settings are available, such as the number of harmonics, the amount of gain reduction for each harmonic, and the base frequency of the hum, which can be set to the preset 50 or 60 hz – the frequency of the DC current in most (if not all) countries, and thus the most likely hums to be introduced during a recording powered by electricity – or freely, should you encounter a situation with a different kind of hum such as standing waves in an acoustically untreated room. Or, as in this case, where the hum emanates from another source of electromagnetic interference than a plain old wall socket.
Using the zoom tools and the ”Play Frequency Selection” button, I pretty quickly zoomed in on the trouble areas to be two pretty close frequencies in the 4,5 – 5 khz area. Experimenting a bit with the settings in the Hum Removal module, I managed to suppress the buzzing frequencies without doing too much damage to the material that I wished to preserve. Thanks to the high level of tweakability, allowing me to set all parameters I could wish for and quite a bit that I had no idea that I might need to do anything to, the resulting audio became this:

music_box_2_hum_removal.mp3

Not that bad results, really.

Next up, I wanted to do something to the slight distortion that may or may not be apparent depending on your listening equipment, mostly prominent around 16 – 20 seconds. The file in itself wasn't exactly overloaded, but because of the low quality of the membrane of the microphone used for the recording, an overload had taken place at an earlier stage than the A/D.
The first and obvious choice to deal with this is the Declipper module. As a result of the issues mentioned above, we are here dealing with a file where the clipping occurs at a level a lot lower than 0 db. This means, that in order for the Declipper module to be able to do it's job, the file needs to be normalized; not a big problem, but I look in vain for a way to do that within RX. I end up having to open the file in another app to do that one task before being able to proceed with this process; adding a ”normalize” option within the gain process window in the next update looks like a no-brainer to me.
Still, the Declipper module presents various ways of doing this, with varying results. Some of these results are of a nature that might invite me to use them creatively, but that doesn't actually remove any clipping, though they might work more succesfully on different program material. The settings that I end up with, however, produce this result:

music_box_3_declipping.mp3

Just to try a different approach, I go on to open the Declicking module. Originally intended for removing vinyl crackles, digital sync error induced clicks and the like, my thinking is that anything intended for getting rid of short bursts of noise might come in handy in this case. The result is different, but not unusable either:

music_box_4_declicking.mp3

Just for the hell of it, I end up trying the Declipper module upon the Declicked audio; again, a different result

music_box_5_declicking_declipper.mp3

Though none of these processes rids the audio completely of distortion, all 3 of them are an improvement over the original. To add, the less succesful results of this particular experiment may very well come down to me not knowing exactly how to use all of the features in these modules, that are customizable beyond what I've seen elsewhere. I urge others needing these processes to downolad the demo of RX to see if they have better skills than me in getting the wanted results.

Now, the audio is slowly but surely getting there, but a few radical steps are needed to get this to an acceptable state. Next up, I'm going to deal with the unwanted noises of the electromagnetic charges causing the 2 alternating tones that can be heard throughout, as well as the chirping sounds of the mechanics when turning the handle of the music box.
I try out quite a few things; the Declicker won't get rid of the hipitched mechanic noises, though you can hardly blame it, since it wasn't meant for detecting and eiminating that kind of audio. The Hum Removal is unsuccesful with the tones, even if they are completely consistent in pitch – but then again, one can hardly call it a ”hum”. And, trying out the Denoiser module (which we'll get to in a bit) will only yield usable results if it gets to eat away a healthy bite of the audio that I wish to preserve too.
In fact, I end up with a tool that's available to every other spectral editor; I simply select the bites of audio one by one, and hit the shortcut key for ”Silence selection”. The frequencies of the tones are quite easy to spot in the spectrogram, and it is quite obvious that these have such a narrow frequence span, that it is worth a try to simply get rid of these frequencies altogether. This turns out to work just fine. After that, it's a matter of going through the audio from the beginning, highlight a ”chirp”, silence, highlight the next ”chirp”, silence, a.s.o., until reaching the end of the file. The end result is this:

music_box_6_spectral_edit.mp3

Even if this process does make the upper frequencies of the spectrogram look like a swiss cheese, this very effectively eliminates the attention drawing hi-pitched squeaks. It does leave the lower frequencies of the mechanic noises behind, but if removing them, this suddenly causes too many unnatural sounding gaps in the audio.

Upon reviewing this text, Izotope points out that the Spectral Repair module, which we'll get to in a bit, has been specifically designed for tasks like this, and would have suited the job better, since it could have repaired the holes left by the ”chirps”, rather than just deleted both the unwanted audio and the ambience surrounding it. Fair point – so, let's move on to the Spectral Repair module:

Normally, I wouldn't have a problem with the current level of mechanical noise in a recording like this. The mechanics are, after all, an integral part of the sound of a music box. But just for the purpose of checking how much can be done with RX, I decide to try and eliminate the lower pitched mechanic noises as well. For this, I'm using the Spectral Repair module.
The Spectral Repair module deals with a selected span of audio by analyzing the audio right before and/or after it (the ”and/or” refers to the fact that the amount of ”before” and ”after” is adjustable), to resynthesize and replace the audio in question. Say, you have an audio file with a 300 ms gap in it; Spectral Repair fills the gap with a resynthesis of the audio surrounding the gap. Or, if you have a long held note in your recording, disturbed by someone in the audience coughing loudly or a door slamming in the background; use Spectral Repair to replace the piece of audio with the cough in it with a resynthesis of the held note that you wanted the listener to enjoy.
In this case, I want to replace the pieces of audio with audible mechanical noise in it with a resynthesis of the bells sounding just before and after the noise. The settings that provide the best results in this case differs from one instance of mechanic noise to the next, and sometimes results in long, CPU demanding processes. The whole cleaning operation on a fairly polluted 30 seconds of audio like this took me some 90 minutes, going from one ”clank” to the next. The result sounds like this:

music_box_7_spectral_repair.mp3

To finish, I'm going to deal with the background ambience hiss of the recording. For this I reach for the Denoiser module, which is a traditional noise profiling tool. This works in the way, that you select a piece of audio made up entirely of the kind of noise that you want to get rid of, asks the Denoiser to analyze this pure noise, then select the rest of the audio and have the Denoiser sort of ”subtract” the noise from the audio through a not completely transparent process.
That's all well; noise profiling is not news, it has been around for quite some years, and is available in audio editors and plugins ranging from the freeware Audacity to Adobe Audition, just to mention the two noise profiling options available on the laptop that this review has been written on. Given that the process of noise profiling relies heavily not only on selecting the right piece of audio for the noise profile (which is of course crucial), but also on the tweakability of the noise removal process, the success of noise profiling is very much dependant on how much one can finetune the process by tweaking parameters, and how well the process responds to these tweaks. As for tweakable parameters, RX's Denoiser has plenty of them, and, as the argument goes here, that is a good thing. Especially the opportunity to make settings for smoothing out the artefacts that noise reduction usually introduces proves to be very effective. One can always discuss the layout of the user interface, where you have to click not one, but two buttons to get the full range of parameters visible, but that's a minor detail. The process responds quite drastically to parameter tweaks, so there might be quite a few hit and miss situations while trying to achieve the desired results. Also, if choosing some of the more extreme settings, processing time rises dramatically – processing these 37 seconds of stereo audio at 44/16 resolution took 3 minutes on a current macbook pro – so count on using quite some time to get to know this module thoroughly.
No more delaying it; here's the result of the Denoiser's work on this recording:

music_box_8_denoiser.mp3

With this step, I've gotten the original recording as close to a usable track for further work as it can get with RX Advanced and the skills I have acquired with the app so far.
Izotope comments on this chapter by pointing out how and why the noise reduction process supplied by the Denoiser Module is way more advanced than the competition. One should always take such sales blurbs with a grain of salt, but you can judge the claim for yourself over at http://www.izotope.com/products/audio/rx/.

Conclusion
RX Advanced offers a set of very powerful tools for audio restoration. As I explained in the beginning, none of the processes presents any revolutions (with the possible exception of the Spectral Repair module), so the selling point will have to be that RX performs better than its competition. I can't claim to have tried every noise reduction software tool on the market, but from my limited experience, RX offers enough tweaking possibilities to cater for the situations where I have needed audio restoration, and it does so with the least amount of audible artefacts that I have experienced. The GUI is intuitive and easy to use, and offers convenient and easy-to-use possibilities for comparison of different settings. The documentation, which is completely electronic, is adequate, though a few chapters explaining the relevant background theory of digital audio would be a welcome addition.

With RX Advanced, Izotope obviously aims at the professional user. This can be seen in the quality of the processes, the help files that take basic knowledge on the users part for granted, and in the way that this does not attempt to be a catch-all application, but rather a tool with some very specific uses. It only does audio restoration, but it does that part well. It can also be seen on the price tag; again, I'm not the most experienced person when it comes to buying tools for professional studios at the level that this software title is aiming at, but it does seem to me that $1199 is quite a steep price. Especially since I can get tools that do these jobs satisfactorily for me for quite a bit less. Still, I urge anyone who does not necessarily find this price point to be too high, and who might need more than what the included tools in standard wave editors will supply, to go and download the demo of RX Advanced. It might be just the right tool for you.