Getting Better Results From Izotope's RX2

Restoration software such as Izotope's RX2 can breathe new life into damaged audio — with the right moves from the user!

Izotope's RX2 software makes available some of the most powerful restoration tools around, at an affordable price, and as a result has become a very popular package. In this article, I'm going to share some power-user hints and tips that will help you get the best from it.

If you're new to RX, a good place to start would be checking out Sound On Sound's original review in July 2008 (/sos/jul08/articles/izotoperx.htm). Since that review was published, Izotope have released a new and significantly expanded version called RX Advanced, in addition to the basic RX. RX Advanced has a number of extra modules, and some of the modules that appear in both versions have extra features in the Advanced release. Both variants can be run as stand-alone applications or as plug-ins for your favourite Mac OS or Windows DAW: Audio Units, VST, RTAS and AAX Native plug-in formats are supported. However, running RX as a plug-in means that its processes have to operate in real time. This makes some features unavailable and limits the effectiveness of others, so I tend to export files that need processing to the stand-alone version of RX and re-import them into my DAW once processed. (For this reason, my number one feature request would be for Izotope to improve the links between DAWs and the stand-alone application, perhaps in the same way as Synchro Arts have done with Revoice Pro.)

The basic version of RX has five main restoration modules. Declip is for repairing clipped and distorted audio, while the Declick & Decrackle module is intended for restoring recordings from vinyl records, although it is also good for dealing with digital clicks. Remove Hum can eliminate low-frequency noise such as mains hum, along with up to seven harmonics. Denoise removes broadband noise that is relatively static in profile; it is effective both on electrical noise such as hiss, and acoustic noise such as air-conditioning. Finally, Spectral Repair is designed to remove occasional random sounds that have interrupted a recording, whether these come from the instrument being recorded or from external sources. Beeps, car horns, mic pops and mouth clicks are all grist to its mill.

Supplementing these are a selection of 'fix it' modules such as Gain & Fades and Channel Ops, which can help with all kinds of routing and phase-related problems. There is also a Spectrum Analyser module to help track down exactly where the problems are. In RX Advanced, there are more 'fix it' modules you wouldn't necessarily expect to find in a restoration software package, such as Resample for downsampling audio files, Dither for reducing the word length of audio files, and pitch-shifting and time correction using Izotope's Radius technology. The Advanced version also operates as a VST and AU plug-in host in stand-alone mode, and boasts an intriguing Deconstruct module, where you can separate and adjust the tonal (pitched) and broadband (unpitched) elements of an audio file.

To illustrate some of the techniques involved in getting the best from the various RX modules, I'll work through some examples of audio files that have needed some work. I will describe how things happen in the stand alone application, but most of what I am covering could be undertaken in the plug-ins within the limits of real-time processing.

Peak Distortion

Digital audio is not forgiving when it comes to peak distortion: if your signal exceeds 0dBFS, you will experience clipping, with anharmonic distortion that is usually very obvious and unpleasant. My first screenshot1: The stand-alone RX Advanced. A stereo audio file is loaded that has severe clipping distortion. shows an audio file where the tops of the waveform have been chopped off, and the challenge is to restore this audio to its former glory using the Declip module. Either adjust the Clipping threshold by eye so that either the red lines on the audio waveform are below the clipped-off peaks, or click the Compute button in the module window and adjust the Clipping threshold control until the red line is just before the white line in the module window display. Next, adjust the Makeup gain. This is actually an attenuation control, and it is needed because once RX has reconstructed the missing peaks on the audio, its new peak level will be around 6dB higher — hence the default setting of -6dB. If you are already close to 0dBFS, you might want to consider reducing this further.2: RX's Declip module. 'Makeup gain' is a misnomer, as it's actually attenuation that is required following the de-clipping process. When you are ready, you can Preview the changes, and if you're happy, choose Process and then Save As in the stand-alone RX to save a new version of your audio file. The third screen shows the audio 'before and after' de-clipping: I have processed only the right channel of a stereo file, and as you can see, its flat-topped appearance is gone, and its average level is much lower.3: Here, you can see what the de-clipping process has done to the right-hand (lower) channel of the audio file.

Hum & Buzz

Screenshot 44: RX's analyser window displaying a mono audio recording with a severe hum problem. As you can see from the intensity of the yellow line across the bottom, the hum is pretty much the loudest thing on the recording! shows a section from a recording where the screen had become disconnected on the mic lead. Because it was recorded in a hurry, the setup wasn't checked, and as it was a recording of a funeral, there was no chance for a retake. You can see that the hum and buzz are nearly as loud as the speech, presenting quite a challenge for RX's Remove Hum module! Before we start, note that there are display options that can help make this kind of problem easier to pinpoint. By right-clicking on the display and selecting View Spectrogram Settings, you can change from the default Auto-Adjustable SFST to 'Adaptively sparse'. As you can see in screenshot 55: In this case, changing the Spectrogram Setting to 'Adaptively sparse' has made the hum and its harmonics easier to pinpoint visually., this makes the hum and its harmonics much clearer on the spectrogram display.

Once you've achieved the best visual settings for your material, open the Remove Hum module. There are presets available for removing common problems such as 50Hz or 60Hz hum, but although these can be good starting points, check on the Spectrum Analyser what harmonics are present: there is no point in using a preset that has deep notch filters for harmonics if there isn't anything there to filter! In screenshot 66: RX's spectrum analyser clearly shows the strength of the 50Hz hum that has been recorded, but also makes clear that the lower harmonics are not present — notch filters will be needed at 250Hz (the fifth harmonic) and 350Hz (the seventh) but not at 100Hz, 150Hz or 200Hz. you can clearly see the main 50Hz peak, but you can also see that no secondary peak occurs until 250Hz — the fifth harmonic of the hum frequency — and, after that, the seventh harmonic at 350Hz. With the 50Hz hum preset loaded in the Remove Hum module, I can deal with this by unlinking the harmonic filters in the Linking Type drop-down menu and then adjusting the second, third, fourth and sixth harmonic filters so they are not cutting any audio.7: RX's Remove Hum module. Here you can see that Linking Type has been set to 'none', and that only the notch filters relating to the fundamental, fifth and seventh harmonics have been activated. Once I click Process to apply the filters to the complete file, you can see from screenshot 8 that the major hum problems are gone.8: After a single pass with RX's Remove Hum module, the bright line along the bottom of the spectrogram is gone, indicating that the fundamental frequency of the hum is no longer present. Its fifth and seventh harmonics at 250 and 350 Hz have likewise been removed, but higher harmonics are still visible.

However, the Remove Hum module provides up to eight filters, for the fundamental and the first seven harmonics — and what is also obvious from screenshot 8 is that the recording contains plenty of harmonics higher than the seventh! These can be dealt with in subsequent passes by using the Remove Hum module's Free mode, configured to supply a single notch filter as in screenshot 99: Once I've tracked down rogue higher harmonics using the spectrum analyser (above) they can be eliminated one at a time by setting up a single notch filter in the Remove Hum module's Free mode.. Here, I've used the spectrum analyser to find the higher harmonics manually and process them one at a time, starting with the ninth harmonic until, as you can see in screenshot 10, the stronger lines have been taken care of.10: The same audio file after a number of individual passes of the Remove Hum module to attenuate harmonics of the hum.

The law of diminishing returns begins to apply as you go higher up the spectrum, but it is usually worth dealing with the more prominent upper harmonics before moving on to the next stage, which is to use the Denoise module to clean up the remaining low-level buzz. Highlighting a section of noise, as shown in screenshot 11,11: Here I'm highlighting a section of noise (with no wanted audio in it) to create a noise fingerprint for the Denoise module. gives the Denoise module a profile of the noise to learn. By going into the Advanced tab (which, confusingly, is available in both the standard and Advanced RX Denoise modules!), it is possible to set separate Threshold and Noise Reduction parameters for the Tonal (pitched) and Broadband (unpitched) elements of the audio, as you can see in screenshot 12.12: The Denoise module's Advanced tab, showing the noise profile captured in screen 11. Setting the noise-reduction parameters separately for the tonal and broadband elements of the signal can help get the best results. This screen also shows strong peaks in the noise profile around 7kHz and 15kHz, and with the Advanced version of RX it is possible to get the Denoise module to target specific areas of the frequency spectrum, using the Noise Suppression Envelope, which is the grey straight line on the display in screenshot 13.13: RX Advanced's Denoise module has the ability to target noise reduction to specific frequency regions, in this case the peaks at 7 and 15 kHz. Creating anchors lets you make the Denoise module reduce certain frequency bands more than others. The problems at 7kHz and 15kHz are clear on the noise profile, so we can use the suppression envelope to target them. Now, when the file is processed, virtually all the buzz harmonics and other noise problems are gone, as you can see in screenshot 14.14: The final result — my problematic audio file has been fully cleaned up with both hum removal and broadband noise reduction.

Nuisance Texts

My next example illustrates another real-world problem that requires work with RX: an interviewee's mobile phone has bleeped a text alert in the middle of an interview that can't be re-recorded. Screenshot 1515: The pitched elements of a mobile phone text alert in an interview are visible as horizontal lines to the right of the cursor. shows the spectrogram of the recording, with different tones in the alert sound visible as horizontal lines — the first one starts where the cursor is. These pitched elements are, again, best tackled using the Remove Hum module in Free mode, where you can manually set the frequency of the notch filter. The problem frequencies are not as obvious as in the previous example, so it can help to zoom in a little more, especially on the frequency axis, and hide the superimposed waveform. Another way to identify problem frequencies in a case like this is to place the cursor in an area that is reasonably clear of other sounds and open the Spectrum Analyser module. You can see from screenshot 1616: The Spectrum Analyser makes it easy to pinpoint the problem frequencies in the text alert. that one of the problems is at 586.4Hz. As the text alert has only a short duration, I've highlighted the problem area on the Spectrogram, so I'm only processing the relevant section rather than the entire file. Since I know the problem frequency, I can type it into the Remove Hum module and hit Process. I can then repeat these steps to remove each one of the individual notes and its harmonics, until all that is left of the text alert are the attack sounds.

To resolve these, I need to switch to the Spectral Repair module and use the Replace tab. I can highlight the offending areas using the Paintbrush or Lasso tools and then click Process to replace the selected area with material from the surrounding area, which is set by the Surrounding Length control in the Spectral Repair window. With care and a bit of trial and error, it's possible to remove this sort of incidental noise pretty well completely, but don't expect to be an expert first time. It does take some experience to interpret the display with the combination of listening and trying different settings, but with a full undo history available in the bottom right-hand corner of the window you can always go back and try again.

Camera Click

Spectral Repair can also sometimes help to tackle a common problem with wedding video recordings, where the microphone has picked up camera shutter noise from a photographer taking still photos. In screenshot 17,17: Here, I have used the Lasso tool to highlight an unwanted camera click in the audio from a wedding video. you can see that the first camera click is preceded by some wanted audio but followed by a period of relative silence. I have used the Lasso tool to draw around it, after which I navigate to the Spectral Repair module's Replace tab and set the Before/After weighting to completely 'After', so that the algorithm draws the replacement material entirely from the 'dead' area after the click rather than the wanted audio before it.18: The Replace option takes material either from before or after the highlighted section. In this case I want to weight it so as to take everything from afterwards. I click Process and the click is gone!

The click shown in screenshot 19,19: This camera click doesn't have any 'dead' space on either side of it, so the standard Replace option won't work too well... however, is almost embedded in the speech, so the simple Replace option won't work too well. Instead, I need to copy and paste replacement audio from further on in the file. To do so, I first draw around the shutter sound, then drag the selection along to a suitably quiet region, as shown in screen 20.20: ...But fortunately, I can copy and paste suitable material from nearby. I can then copy this region into the clipboard, drag the selection back to the shutter sound, and paste the clipboard audio over it.

Vinyl Record

Finally, let's see how to use the Declick & Decrackle module to remove scratches and surface noise from a vinyl transfer. The Declick tab provides three algorithms to choose from. Single-band is really designed for noises of very short duration, such as digital clicks, but there are two multi-band options, one for repetitive clicks such as long scratches, and the other for random clicks. Whichever you use, the way to get the best from it is to select Output Clicks Only and play your audio in preview with the Strength control at maximum.Screen 21: The Declick & Decrackle module features separate algorithms designed for digital clicks and periodic or random surface noise. You will hear that the Declick module outputs some of your wanted audio transients along with the clicks; gradually reduce the Strength until all you hear is the clicks. When you are happy, de-select the Output Clicks Only option, click Process, and your clicks will be gone.

I hope these examples will help you get the most out of Izotope's RX2 restoration software. And if you don't own it but this article has piqued your curiosity, you'll be pleased to know you can download a trial version: click on the Try button on the RX2 page of Izotope's web site (http://izotope.com/products/audio/rx/).