Spectral Dynamics – Dynamic Equalization, Part Three Continued
OK, now that you’re familiar with most of the controls and what they do, we can get to the fun part: what you can do with this amazing and versatile plugin. This is by no means an exhaustive list of things you can do with it, only some of the stuff I’ve had the best success with. Hopefully, though, it’ll inspire you to explore other, perhaps more creative and less-obvious applications. I’m finding new applications all the time.
Usage Note #1: Mastering
The easiest application, and perhaps biggest bang for the buck is simply inserting MSpectralDynamics in front of the limiter on the master bus. Use it lightly, much as you’d use a normal “glue” 2-bus compressor to help the mix gel and/or maximize volume.
I usually start out with fast attack and release times, although whether that’s appropriate or not depends greatly on the material. For up-tempo, aggressive wall-of-sound music a 1ms attack and 10ms release might work just fine. For a quiet ballad, much longer attack and release times will probably do better.
I’ve gotten the best results keeping the threshold all the way down at its minimum, and the ratio between 1.1 and 1.25. Experiments with higher ratios and higher thresholds haven’t produced results as good as low ratios, but feel free to experiment.
I adjust the Smoothness parameter based on mix density: 2% for less-dense material such as acoustic guitar with vocal, 1% or disabled entirely (0%) for very busy mixes with many tracks. The lower the Smoothness value, the more important the Naturality control is, so if I drop Smoothness all the way down to zero, I’ll often set the Naturality knob to 2% or 3%. Otherwise, Naturality normally stays at either 1% or 0% (off).
If you find that the plugin is compressing the low end too much, use a custom threshold curve to bring the low end back up. Two nodes is all it takes, one at around 120 Hz and another around 200-240 Hz, as show below. Then raise the 0-120 segment until the bass comes back. Note how the compression indicator shows less compression in the low end.
As the final step, back the Wet/Dry mix knob down until you’re using the minimum amount of compression to do the job. I usually end up around 80%, but it depends on the mix. Generally, the better the mix the less you’ll need, so I guess my 80% average says something about my mixing abilities!
Usage Note #2: Vocal Smoothing
Regardless of the genre, smooth and level vocals are expected. Whether they’re whispered, crooned, screamed or grunted, sweet or nasty, listeners want them to be consistent in both volume and tone. That’s why vocals are almost universally compressed (unless you’re working with world-class pros and high-end microphones in an ideal acoustical environment, but that’s certainly not my situation!)
Spectral dynamics processing will almost always benefit vocals, no matter what the style. It’ll smooth out little inconsistencies caused by cheap microphones, poor microphone technique and room resonances, and make harmonies blend better. You can insert the processor on the vocal bus, or use separate instances for lead and background vocals. I prefer to treat lead and background vocals separately because I like to compress BGVs more severely than the lead, and use shorter attacks and longer releases.
Start with a fast attack and then gradually lengthen it if you’re losing your “T”’s and “K”’s. At most, I’ll use up to 10ms for attack and almost never go higher than that because those “T”s and “K”s are only 4 to 7 milliseconds in duration.
Here’s a tip: if you’re doing background oohs and aahs, run them all to a bus and smooth them out with spectral dynamics. Level-automate and compress them individually first, and then let MSpectralDynamics supply the final sheen. Use a little higher ratio, perhaps up to 2:1, a fast attack (1ms) and long release (150-250ms).
Also, don’t forget about the gate for suppressing breath sounds and other inter-phrase noises.
Usage Note #3: Sibilance and Plosive Control
MSpectralDynamics makes a pretty good de-esser. Perhaps not quite on par with Fabfilter Pro-DS, but it’s easily as good as or better than your garden-variety de-esser. It’s also very easy to dial in, because you don’t have to search around for the sibilance frequencies. The plugin will figure that out for you.
Surprisingly, there is no bundled de-esser preset. That’s not a problem, though, as it’s pretty easy to set up. All you need is a custom threshold curve with a low threshold within the sibilance range, from about 4KHz to 10KHz. The Analyzer section will allow you to dial it in more precisely.
Here’s the basic setup: Resolution: 10ms, Smoothness: 0%, Naturality: 0%, Slope: 0dB, Attack: 1ms, Release: 10ms, Ratio: 2:1. Everything else can start out at default values.
Next, we’ll set up the threshold curve.
Click on the Threshold button to enable editing, and then add two nodes, as shown in the image on the right. I’ve placed the first node around 3KHz and the second at around 4KHz.
Don’t worry about being precise with the node frequencies or threshold, as we’ll dial these in during playback. Initially, all we need is a custom threshold like this one.
Next, play back the vocal track and observe where the sibilance falls in the Analyzer display. As the image on the right shows, it’s pretty easy to see where the sibilance lives.
If you observe that your node frequencies are way below the sibilance region, adjust them upward so that you’ll be affecting the narrowest range of frequencies possible and thereby assure that the process will be transparent. But precision isn’t all that important, because the plugin is smart enough to only compress those frequencies that exceed the threshold.
The final step is to listen to the vocal while moving the right-hand line segment up and down. You’ll know when it’s too far down because the “S”s will start sounding lispy.
The same procedure works for reducing plosives, low-frequency pops caused by “P”s and “B”s and the big burst of air they can shoot into the microphone’s diaphragm. It’s the same procedure, but with the threshold lowest down in the 100Hz and below region where plosives live.
Now, you may be asking why this is better than a conventional multi-band compressor or a broadband compressor with a sidechain filter. It’s better because compression is driven by an ongoing analysis of the sibilant frequencies, wherever they happen to fall and whatever their amplitudes. This is great when sibilance doesn’t neatly fall into one well-defined band, but instead pops up in different bands at different times, or varies greatly in amplitude.
There may come a time when you’re presented with a vocal track that’s so ess-ified that it defies attempts to submit to any de-essing plugin. This happened to me once when a ribbon microphone turned out to have a loose element inside that caused a high-frequency resonance that would make your ears bleed, which I’d failed to noticed during tracking because I was the singer. For those especially difficult de-essing situations, clone the vocal track, compress it heavily and apply a deep high-pass filter, and then run that into MSpectralDynamics’ sidechain input.
Usage Note #4: Noise Removal
Noise removal is the process of analyzing the spectral footprint of a noise component and then reducing just those frequencies whenever they are detected. Many audio editors have excellent noise-removal features, and there are de-noiser products available such as iZotope’s amazing RX3. But if you don’t have such a dedicated solution (RX3 lists for $1200 – it’s not in my kit), then you can turn to MSpectralDynamics to pull this off.
It’s not a general-purpose noise-removal tool and it won’t work in all circumstances. But when it works it works quite well, and the technique is worth mentioning if for no other reason than to show just how versatile the plugin is, and perhaps to get you thinking about other novel or off-label applications that can take advantage of the spectral detector.
To start, you’ll have to find a piece of audio that just has the noise and nothing else. That’s necessary because MSpectralDynamics needs to be able to analyze it in isolation so it can learn what to remove. All de-noisers work this way. Ideally, the noise sample will be at least a second or two in duration for best results, but you can often get away with a much shorter noise sample if that’s all you’ve got. Some engineers record a few seconds of quiet room sound just for this purpose, a standard practice for audio recording on movie sets.
There is a handy de-noiser preset built into the plugin. Go into Easy Mode if necessary (click on the Edit button to toggle between easy and edit modes) and select the “Denoiser” preset from the list. After selecting the preset, click on the Edit button to go back into Edit Mode.
Since the noise sample will probably be short, set your DAW to loop on just the noise. Start playback and click on the Capture button above the Analyzer window. If the noise is very quiet, you’ll probably have to turn up the Input Gain. You’ll know when you’ve got a good capture when the spectrum shows in the Analyzer window, as shown to the right.
The plugin has now acquired a spectral profile of the noise that will allow it to distinguish the noise from the non-noise signal.
If you’re not seeing this, turn up the Input Gain until you do.
Now all we have to do is disable the loop, play back the track and move the threshold knob up and down to find the optimum amount of noise reduction.
Start with the threshold all the way at the top and slowly lower it while listening to the playback. As you move the threshold down, at some point you’ll hear the noise start to disappear. Keep going until you start to hear a tonal change in the main audio, which lets you know when you’ve gone too far. Bring the threshold back up until there is no audible effect to the main audio. Bypass the plugin to make sure you’re not altering the track’s tone.
Just how much noise reduction you can achieve will depend on the track, how consistent the noise is, and how spectrally different the main audio and the noise are from one another. The good news is that it works great on computer fan noise and AC hum, the two most common problems in home studios. It would probably work well on other steady hissy noises such as air-conditioners, but here in the cool Pacific Northwest of North America air conditioning doesn’t exist, so I can claim no personal experience with that.
Usage Note #5: Spectral Ducking
I’ve saved the best for last. Don’t bother googling “spectral ducking”. I just made that up. But in lieu of any other accepted term, it’ll have to do.
First, a general definition of ducking for readers who aren’t familiar with the concept. It means lowering one signal proportionately to some other signal. The classic example is a radio commercial where a bed of music dips in volume whenever the announcer talks over it. That’s known as “ducking”.
Any compressor can be a ducker, as long as it has an external sidechain input. “Sidechain” refers to the signal whose amplitude is being measured to determine how much gain reduction to apply. Most of the time, the sidechain signal is the same signal we’re compressing, in which case we say we’re using the “internal sidechain”. If we instead use some other signal, from another track or bus, then we say we’re using an “external sidechain”. Ducking therefore requires an external sidechain.
Here’s where it gets interesting. MSpectralDynamics isn’t just measuring overall amplitude like a normal compressor would. Instead, it’s continually analyzing the spectral composition of the incoming signal, and applying compression (and/or expansion) based on that analysis. When you send an external sidechain to it, it analyzes that other signal’s spectral composition instead, and applies compression to the main signal accordingly.
The result is that MSpectralDynamics carves out a spectral hole in the signal that is the mirror image of the external sidechain signal. It’s only ducking whatever frequencies are present in the sidechain. If you’re thinking that this might be useful for kick/bass conflict resolution, yup, it does that very well. But that’s a pretty mundane application. Let’s look at something any old sidechain compressor can’t do.
Picture this: MSpectralDynamics inserted into a bus where all (or most) of the mix’s instruments are routed, and the song’s vocals sent to its sidechain input. The plugin is now analyzing the vocal spectrum and using that to compress the instruments. In other words, the instruments are being ducked by the vocal – but only those frequencies that conflict with the vocal are being reduced! As a result the vocals become clearer, while instrumental energy and overall loudness remain constant.
This trick is especially useful for dense rock mixes and genres where the vocal doesn’t sit way up on top of everything else, such as in a pop ballad. It’ll let you keep the bass, drums, dense guitars and synth pads roaring and still have the vocals cut through as if they’d been mixed much louder than they actually were. And because we’re only carving out spectral space when it’s needed, space that’s perfectly filled by the vocal, the effect can be completely transparent. There is no pumping, and no clue that instruments are giving anything up for the vocal.
Of course, you can also use this technique for mitigating any kind of frequency conflicts, making room in the mix any time two instruments are fighting over the same spectral turf. For example, imagine a fat pad or strummed 12-string that fills all the holes like it should – but steps aside when the guitar solo starts. Any time you want one instrument to get out of the way of another instrument, and do it transparently, spectral ducking can be the solution.
Here’s How to Do It
First, enable sidechaining. Click on the Sidechain button at the right-hand side of the Spectrum panel. In sidechain mode, the Slope control has no effect and the Wet/dry control is used to reduce the effect of the sidechain signal.
Detection and compressor controls work the same way as with internal sidechain mode. You can even create a custom threshold for the sidechain signal, as well as a custom processor curve.
- Insert an instance of MSpectralDynamics on the instrument bus
- Route the vocal(s) to its sidechain input
- Set the Dry/wet control to 100%
- Select manual release mode
- Set attack to 40ms, release to 50ms
Now we’re ready to dial it in. Play back the project and listen for the instrument volume change when the vocals start. Ultimately, you’ll want to tweak it in such a way that the compression action is not noticeable. The best way to do that is to start out with the effect exaggerated to the point where it’s easily heard. This will make it easier to dial in the best attack and release times.
Start by adjusting the attack time. The longer the attack you can get away with, the better. The idea is to have compression begin slightly after the vocal starts. If you’re missing too much of the beginnings of vocal phrases, then your attack time is too long. If the transition into ducking is too abrupt, then your attack time is too short.
Setting the release time is slightly trickier. Vocal phrases often decay on the last word, so setting the release too fast can result in losing the tail-ends of phrases. Setting it too slow will cause an audible gap between the end of the vocal phrase and the instruments coming back up. Release times that have worked for me have ranged from 10ms to 500ms, depending on the nature of the vocal phrasing, so it really needs to be adjusted by ear. Start at 50ms and gradually increase it until you hear a natural fade-up. Make sure you audition multiple vocal phrases, because the ideal release for one phrase might not work for another.
Once you have the attack and release times delivering a smooth transition, start turning the Wet/dry control down until the instrument volume pumping goes away. This can end up anywhere from 30% to 80%, depending on the compression ratio and sidechain level.
When done correctly, listening to playback should give no clue of the trickery you’ve just done. But bypass MSpectralDynamics and the difference will be obvious: the vocal will sink into the mix and not sound loud enough. Spectral ducking should let you mix the vocal 3 to 6dB quieter than normal and still cut through. And that means the instruments can be 3 to 6 dB louder and no longer have to lay back to accommodate the vocal.
If, in the course of reading this piece, you began to suspect that I am a less than impartial reviewer, then you’re right; I absolutely love this plugin!
Despite being an admitted plugin enthusiast, over the years I’ve steadily whittled down the number of plugins that I regularly use. My belief is that every plugin you own does more than you think, and it’s better to deeply learn what you’ve got before blowing more money on the next glittering gem. So these days there aren’t many plugins that find their way into almost every project. MSpectralDynamics is one of them.
It’s not a big product category, but there are a few other plugins out there that perform in a similar fashion to MSpectralDynamics, or at least perform some of the same tricks.
Soniformer from Voxengo ($89.95) lacks the automatic intelligence, but for a manual solution it’s quite handy, especially for resonance remediation and volume maximizing. It actually outperforms MSpectralDynamics in some situations, such as when aggressive compression is called for.
Dynamic Spectrum Mapper from Pro Audio DSP ($329) comes closest to MSpectralDynamics in functionality and adds a few tricks of its own. Paul White wrote an excellent review of this plugin in 2009 for Sound on Sound, at which time DSM was a Mac-only product. Since that article was written, DSM has been ported to Windows as well.
Wavesfactory’s Trackspacer (59 EUR) isn’t a general-purpose spectral processor, but it does perform the spectral ducking trick described above and is inexpensive if that’s the only trick you’re after.
Space Boy from Elevayta ($29.95) is also dedicated to spectral ducking plus has the ability to identify and display frequency conflicts. It’s a bargain at $29.95 but unfortunately for Mac users, it’s Windows-only.
For Noise Removal
There are a great many more options when it comes to noise removal.
Topping the list of audio restoration and repair software would be iZotope’s excellent RX4 [LINK: https://www.izotope.com/en/products/audio-repair/rx/ ] ($349 – $1200). Expensive, but covers a lot of bases and might be the only audio repair software you’ll ever need.
WaveArts’ Master Restoration Suite ($349.95) offers noise reduction and click/pop/hum removal. It’s a collection of four plugins, each of which can be purchased separately for between $49 and $299 each. The noise reduction component is called MR Noise ($299).The company also has a plugin called Dialog ($249.95), a channel strip specifically for dialog processing, including noise removal.
Sony Noise Reduction 2 ($279.95) not only does steady-state noise reduction but also pop and click removal. This is a DirectX suite, so it’s a Windows-only solution.
Sony Sound Forge ($300-$600) is a multi-purpose audio editor that also includes noise reduction tools. It’s overkill for just the noise reduction, but it does a great deal more than that.
And let’s not forget Audacity, the best bargain around for audio editing – it’s free! Audacity includes a noise-reduction feature that works much like MSpectralDynamics, and it can be as effective with steady-state noise profiles. It’s a handy tool to have around for general editing tasks and MP3 conversion/tagging. Being free, there’s no reason not to add it to your arsenal.
Download the demo here: http://www.meldaproduction.com/download.php?plugin=MSpectralDynamics
Get the manual here: http://www.meldaproduction.com/download/mspectraldynamics_documentation.pdf
Introduction to MSpectralDynamics: https://www.youtube.com/watch?v=63WibGWENDE
Practical examples: https://www.youtube.com/watch?v=y00kbJBLRFc
Noise reduction: https://www.youtube.com/watch?v=V_K4V_5Kth0