This new trio of plug-ins from UK-based Sonnox (the group that brought us plug-ins based on the Sony Oxford digital console) includes Oxford DeClicker, Oxford DeBuzzer and Oxford DeNoiser. They represent over a year and a half of research and development, and promise significant advances over competing products.
I’ve been using restoration software professionally for about 15 years. During this period, software companies made big advances in transparency and effectiveness. In recent years, improvements have come in smaller increments, leaving many of us wondering, “What’s next?” This situation has not gone unnoticed by Sonnox, who are presenting not just new versions, but a new way of approaching restoration. For the first time, we’re seeing intelligence and processing discretion being built into the applications. For example, many contemporary restoration applications work by examining a section of the audio, then processing the whole file. It is sort of like a guided missile. You program, aim, and let loose on a target. The problem with this approach is that source material changes. In some sections, one type of processing is appropriate, while in other sections, it is not. To get around this, more astute restoration engineers would chop up source material, process each section individually, and then reassemble the finished product. While the approach works, it is labor and time intensive, can cause cross fade problems when the audio is stitched together, and is prone to human error when working with extended duration materials ( e.g., classical recordings). Large restoration projects take much longer in time and more in fees. The new approach from Sonnox hopes to address many of these concerns.
DeClicker helps to remove clicks, pops, and other quick-burst disturbances. There are three different sections devoted to managing pops, clicks, and crackles. Instead of the standard click-as-a-line display used by almost all click-correction software, the main event window displays events as three-dimensional bubbles across the frequency spectrum (the louder the event, the larger the bubble). By lowering the threshold, more events are detected, while the sensitivity control tells DeClicker how many events to repair. A Diff (difference) button allows you to hear the content to be removed, which is crucial for avoiding over-processing. Without this feature, you can easily remove “good” source signal along with “bad” source signal — throwing the proverbial baby out with the bathwater.
One of the most intriguing features is the Exclude Box, which allows you to tell DeClicker to ignore audio by frequency or amplitude. For example, suppose you were processing a drum track but wanted to keep the hi-hats. With other applications, you would have to lower the threshold to allow the hi-hats to pass unprocessed. However, this would allow the loudest clicks through as well. By drawing an exclude box on the area around 12 kHz, for example, DeClicker will ignore the hi-hat information in that vicinity but remove clicks in all other areas.
Dialogue Mode is another example of building pseudo intelligence into the system. It allows you to choose two sets of Threshold/Sensitivity settings: one that is used during speech, and a second for background and non-dialogue audio. Anyone who has restored narration, interviews, or theatrical recordings has already set down this issue and is purchasing Restore. For those still reading, this level of control could save hours and hours of editing and processing time. You’ll have one of those “why didn’t I think of that” moments the first time you see the Dialogue Mode screens.
The second title in the Restore collection is DeBuzzer (think guitar amps and single-coil pickups — bzzzzzzzzz). Most debuzzing applications use a standard comb filter that removes the fundamental and its associated harmonics. For 60 Hz hum (or 50 Hz, depending on where you live), this has been a straightforward process. However, when you run into hums not associated with the electrical grid, this approach falls apart. That’s because it is difficult to find the specific source fundamental.
Sonnox has developed three tools to help users lock in on these troublesome frequencies. First, there is a sweepable Frequency Detect option. Just like it sounds, it allows you to scan the area you suspect contains the problem energy. The second is a large FFT (Fast Fourier Transform) display specifically tuned to find consistent components while averaging out other audio. As long as an instrument or voice isn’t holding a note, the buzz should stand out visually on this display. The third tool is a built-in tone generator. DeBuzzer can generate a control tone that plays along with the audio. As you sweep closer to the source of the buzz, the control tone and the fundamental will have a slow beat frequency. In use, this is very similar to tuning a guitar with harmonics. Once you find the source fundamental, you select it and tell DeBuzzer to attack. It will calculate a comb filter specific to the source buzz and eliminate it. Fortunately, you don’t need to remove buzz in audio very often; that being said, when it’s present, it can ruin an entire track. Having DeBuzzer, especially with its ability to seek and destroy non-standard buzzes, is wonderful.
Reaching into my folder of horrors, I pulled out some single-coil guitar buzz files. Not a problem for DeBuzzer. In fact, the default setting removed the buzz as if it were never there. Moving on, I used a live-sound recording where ground-loop buzz tormented the bass and guitar tracks. With only minimal tweaking to the defaults, DeBuzzer cleaned up the audio. I finally found a recording that had fret buzz on an acoustic. Chopping up the sections in Sequoia, I was able to apply DeBuzzer to the sections that had the problem. Since I couldn’t find a default that addressed this problem, I used the FFT display to get an idea of where the problem was. Then I used the tone generator to hone in on the fundamental. Once I found the root, DeBuzzer did the rest. I wish I had DeBuzzer back when we were working on that record.
Noise reduction is probably the first thing that comes to mind when considering restoration software. There are a few ways to go about noise removal; for example, one popular application uses an approach based on digital imaging. But most applications, DeNoiser included, use a “noise fingerprint” method. This process works by isolating a region where only background noise is present. The software analyzes this noise and removes it from the entire file. Done. In theory, this would be a perfect solution. But in the real world, the noise is rarely static. DeNoiser takes a slightly different route to determining what is noise and what is desirable audio. It examines the entire audio spectrum to determine where the noise occurs and how prevalent it is at different frequencies over time. The “over time” part is the key because DeNoiser evaluates the noise relative to the general signal level. This means that DeNoiser shows discretion as to the amount of noise it reduces. For example, with traditional approaches, areas where there are pauses, rests, or gaps will end up having the source audio removed as the signal level drops to silence. Raising the removal threshold would allow the source audio to stay. The trade off is increased noise level. With DeNoiser, as the source audio dips to the noise floor, only the noise is removed, leaving a smooth decay into silence.
I used a couple of different sources to test DeNoiser. First, I used a live recording done at the University of Pittsburgh’s Bellefield Hall Auditorium with a spaced pair of Audio-Technica AT4050 mics (Tape Op #33) set in omni. They picked up room reflections and the HVAC system. This was a big test for DeNoiser as the air conditioning would cycle at different times. With other noise reduction applications, I would have needed to slice the audio into segments — when the HVAC was on or off. Then, I would have had to process each differently and string them back together. This would have taken a long time for a 70 minute concert. DeNoiser was able to change the level of processing on the fly. The real secret here was using the Diff button, which allows you to hear the audio that will be removed by DeNoiser (all of the Restore titles have this feature). It’s easy to get carried away and overdo noise reduction. With the Diff button enabled, you can adjust the aggressiveness of the software until you start to hear audio content. At that point, simply back off the reduction amount and you are at an optimal “noise-removed-to-signal-untouched” ratio. And while no noise reduction program can work miracles, the results with DeNoiser were very impressive. In fact, average listeners who were unaware of the HVAC issue never noticed it in the final render.
After using the Restore bundle for several months, I have the following thoughts. (1) For many users, simply instantiating the appropriate plug-in will solve your audio woes. The default settings were perfect for a surprisingly wide range of issues. However, if set-and-forget is not enough, the advanced controls provide additional control over processes. Actually, “additional” might not be the correct term — extensive is more appropriate. (2) It’s easy to get lost in the controls. I recommend having a copy of the manual nearby when going into a tweaking mode. There are also about a dozen other interesting features that were not covered here due to space constraints. Be sure to check those out should you demo the software. (3) Finally, at a street price of about $1900, Restore is a significant investment. However, once you realize that Restore goes head-to-head with offerings from CEDAR and other top-notch titles, the price is but a fraction of what the other options would cost.
Mastering and mixing engineers should strongly consider the Restore bundle. It handled some of the toughest tasks I gave it — many times at the default setting. Impressive indeed. ($1895 street; www.sonnoxplugins.com)
–Garrett Haines, www.treelady.com
by Larry Crane
Every stereo mix has a phantom center, or middle channel. This “mid,” includes any audio that is not panned (or recorded) exclusively to the right and left “side” channels, and...