ADX SVC’s simple, user-friendly GUI interfaces with intensive, cloud-based processing to separate speech from background noise. The Separation Options (on the right side of the GUI) affect the quality and content of the extraction.
ADX SVC (Speech Volume Control) is Audionamix’s follow-up to its revolutionary ADX VVC 3 (Vocal Volume Control) plug-in released last year. VVC can separate vocals, or a melodic instrument, from a baked-in mono or stereo mix and discretely adjust its volume and panning. Similarly, SVC separates speech from background noise for discrete volume adjustment of both elements. I was completely blown away by VVC 3 when I reviewed it for Mix’s sister publication, Electronic Musician, so I was especially eager to review the new ADX SVC.
Version 1.0.0 of the cross-platform SVC is available in AAX Native (64- and 32-bit), AU and VST formats. I reviewed the AU plug-in in Digital Performer 9.12, using an 8-core Mac Pro running OS X 10.9.5.
IN THE CLOUD
Like VVC, SVC uses cloud-based processing. Access to the ADX cloud is granted by an API (application programming interface) key you receive after purchasing or renting the software. (iLok is not used.)
After instantiating SVC on a track insert, you make a selection of the audio you wish to be separated (in DP 9.12, selecting the audio is unnecessary). After activating the plug-in’s Acquire button, you either perform an offline bounce or play through the selection so the plug-in can acquire the audio. Subsequently pressing the Separate button sends the data to Audionamix’s ADX servers for processing.
The GUI’s Separation Options should be selected before you press the Separate button. Activating the AVAD button enables Audionamix’s Automatic Voice Activity Detection algorithm during the separation, processing your selected audio only where speech is present instead of throughout. If SVC doesn’t extract speech in some places, turn off the plug-in’s AVAD function and run the separation process again in those spots.
Turning on SVC’s Reverb option separates reverb along with the dry speech so you can maintain the wet/dry balance of the original dialog as you make subsequent changes to its gain. Activating the plugin’s HF Boost option helps SVC extract the high-frequency, noisy components of speech produced by enunciated consonants. Activating the High-Quality button yields better-sounding results at the expense of a longer wait on processing.
To help target the voice for separation, select one of three Speech Settings presets: Male, Female or Child. You can make the separation even more precise by dragging two Pitch Range sliders to delimit the fundamental pitch range of the speech to be separated. After separation is complete, you can drag Speech and Background sliders—and automate them—to adjust the respective volumes of dialog and background noise ±12 dB. Left- and right-channel meters show input levels to the plug-in before separation and output levels afterward.
Even in High-Quality mode, SVC demanded virtually no local CPU resources in DP—no doubt due to all separation processing being carried out in the cloud. High-Quality separation typically took only a little longer to complete than the program length itself. The disadvantage of cloud-based processing is that if your Internet service goes down during a session, you’re out of luck.
I always got better results with the AVAD button turned on; turning it off introduced phase-y, watery-sounding artifacts in any noise or reverb that remained after processing. On an adult female’s dialog track, I turned on SVC’s HF Boost function and deactivated the Reverb button. The track had a steady-state noise floor (from the cheap mic and preamp used by the videographer) only about 26 dB lower than the RMS level of the voice. I boosted the Speech slider around 6 dB—any higher and I would have clipped SVC’s output—and lowered the Background slider 12 dB. The processed dialog sounded clear, unchanged in timbre and fullness, and free of artifacts. Background noise was greatly reduced, though still audible in the lower midrange band (more audible than when I processed the track using iZotope RX5 Advanced’s De-noise plug-in). SVC did not effectively reduce loud clicks from the track.
For grins, I instantiated MOTU’s Trim plugin in the insert before SVC, so I could attenuate SVC’s input enough to allow boosting the Speech slider to the max without clipping. I expected this to screw up SVC unless/until I ran its separation process again, as it had already executed its analysis and processing without Trim upstream in its signal path. Instead, I discovered something that blew my mind and that seemed to defy the laws of science: As I progressively lowered the gain in Trim, only the steadystate background noise level got reduced at SVC’s output. Even with a whopping 40 dB of trim applied, the dry dialog track was just as loud (and pristine). The noise floor, on the other hand, was gone—like in a vacuum! This trick was undocumented at press time because even the folks at Audionamix were unaware of it. (To render this effect, you must route the track out of DP and then back in, recording it to a new track. Simply busing to a new track within DP results in both noise and dialog being attenuated.)
Next up was an adult female’s dialog track that had been recorded outdoors and was sullied with very heavy wind noise. I could attenuate the wind noise at least 7.5 dB without hearing phase-y artifacts. However, SVC’s processing caused the dry dialog’s level to modulate significantly. I achieved roughly the same amount of noise reduction overall using the more expensive iZotope RX5 Advanced Denoise plug-in (which didn’t modulate the dialog level). De-noise didn’t attenuate the sound of strong wind gusts containing heavy low-bass energy as much as SVC, but inserting a rumble filter after De-noise took care of that. Oddly, running two SVC plug-ins in series (in the quest for greater noise reduction) overlaid an out-of-sync, sped-up, chipmunk-like voice over the dialog track on playback.
I next tested SVC on an adult male dialog track that had been recorded in an extremely reverberant room. With the Background slider plunged to -12 dB, SVC significantly reduced reverb and preamp hiss on the track. Recalling the processing I’d used during the original session, however, iZotope RX4 Advanced Denoiser and DeReverb (placed in series) achieved far greater noise and reverb reduction. Of course, RX4 Advanced is many times more expensive (and more complicated to operate) than SVC.
Interestingly, using my pre-attenuation trick (placing MOTU Trim upstream of SVC after processing) yielded poor results when treating tracks polluted with reverb or wind noise: Even minor input-level trimming made SVC sound very phase-y. But as I mentioned before, this unorthodox technique worked like a veritable miracle for reducing preamp noise (noise that’s unwavering in intensity and spectral content).
My only disappointment with SVC’s GUI was that the Background and Speech sliders didn’t show readouts for their current positions when dragging them, making exact adjustments an aggravating task. Otherwise, the GUI was extremely easy to learn and operate.
SVC is best at reducing steady-state, broadband noise. Attenuate its input with a third-party trim plug-in (after completing separation processing), and its effectiveness in that application is downright scary. For engineers on a limited budget, SVC is also a good choice for reducing reverb and wind noise on tracks, though not without a significant trade-off in the latter application. But considering its low price and mind-blowing performance in treating steady-state noise, even demanding pros who own a full suite of high-end post-production and restoration tools should consider buying SVC.
PRODUCT: ADX SVC
PRICE: $199 purchase; $19.99 two-week rental
PROS: Transparently reduces steady-state noise (miraculously so when its input is attenuated after processing). Fairly effective in reducing reverb and wind noise. Inexpensive. Extremely easy to learn and use. Negligible CPU load.
CONS: Ineffective at reducing clicks. Treating wind noise causes dry dialog levels to modulate. Online processing is vulnerable to Internet outages.
Contributing editor Michael Cooper is a recording, mix, mastering and post-production engineer.