Updated: Feb 12
The quest for faster workflows and the use of artificial intelligence in the world of audio production continues. As seen this year at winter NAMM, a plethora of developers have been unleashing their DAW plugin aimed at making the whole production process faster, with features that would be virtually unachievable without today's computational power.
The idea behind this plugin is to work on an already mixed program, and allow the attenuating or boosting of reverb components, as well as changing the relationship between the elements in the mix, allowing producers to fix otherwise unusable material.
The technology is proprietary and labelled as artificial intelligence based MAP (Mixed-Signal Audio Processing) technology. The use cases they list verbatim are:
· Reduction of reverb on location sound and dialog
· Matching of reverb amounts and thus perceived distance on multiple location/dialog recordings
· Reduction of reverb on sub-optimally miked musical recordings, or recordings from venues with less-than-ideal acoustic properties.
· Up-Mixing by separating reverb and direct signal, and using these complementary signals for rear and front channels
· Reducing or boosting “mud” or “grit” in a music mix to bring out the details or move the mix out of focus to make space for additional signals
· Moving sounds/ambience in to or out of focus for On-Screen/Off-Screen differentiation
· Moving background sounds further back to allow foreground sounds to stand out clearer
· Creative sound design
· Generating dynamic “states” from just one recording in game audio.
Installation requires following the activation procedure, which is explained in the PDF supplied with the plugin. It is relatively simple, but requires the creation of a free iLok account if you don't have one. This is because the account will work as a virtual hard lock. I suspect that if you go old school and buy the software package in a shop you will receive a dongle that performs the same function.
All well explained in the PDF manual that unveils when the plugin is installed, the controls are quite diverse and sometimes a bit tricky to understand. Fortunately the PDF manual has some examples of usage in simple language. Since the plugin has plenty of presets, you can also try them out and see what they do with the settings of each control, and go from the one that is closer to your use case.
· Focus and Focus Bias: This control and the associated group of slider settings was the most sensitive during our tests. It's the main control behind the reverb attenuation or boost. Setting at the middle has no effect, lower will add a bit of reverb, and higher will remove reverb.
· Localize: According to the vendor, it "determines pattern localisation along the time and frequency axes", so higher values help with resolution of individual frequencies, but might sound unnatural, whereas lower values may cause "pumping" effects in the signal. During our tests we used higher values when working with a full mix.
· Refract: This control determines the refractory time for the analysis network, so it affects the time the algorithm has to react on what it is hearing. Analogy is made to the attack time on a compressor; so shorter time will make it more sensitive and reactive but might sound unnatural, and longer may cause some of the reverb reflections to return. During our tests we left his pretty much neutral.
· Adaptation: Controls the adaptation time of the analysis network, so a forgetting factor and should be close to what the length of the reverb tail is. It is recommended that you compare the decay curve from both original and output so they look similar. In our case, we used the same settings - quite high - in all of our tests.
· Presence: The explanation is that it adds a random signal to the network's input on each pass of the analysis recursions, with the aim of causing the signal to be a bit more natural. Their manual says that it doesn't really affect any kind of signal, and in our case seemed to make very little difference.
Further controls are detailed in their manual but these are the essential ones we used during our testing.
Testing Part One: Drums
In the first test series, we wanted to understand if this plugin would be able to recover from a bad reverb situation. Since we didn't have any problematic live recordings handy, we've created a track on Logic Pro X with just Drums and bass, but where the Engineer 'accidentally' routed the drums to the auxiliary bus only, as that is where some reverb is applied. Normally you would do a dry and wet mix in parallel just to spice things up with reverb, but in this case we 'forgot' to add the dry signal so the drums reverb are very noticeable and quite unnatural. The bass is just dry, out of the Apple Loops library. We also included the original dry signal clip here for comparison.
We tried a few different presets until we landed on the 'Post ambience de-reverberation' which seemed to do the job we were looking for. Out of the box it sounded a bit unnatural.
We then adjusted the focus frequency sliders to try and match the frequency range of the drum set where the reverb is heard. Following that we tweaked carefully the parameters one by one, until we reached a more pleasant sound. While it did not make the drums sound as dry as in the mix without the issue, we managed to get close enough to something a bit more convincing.
Testing Part Two: Full Mix
For the second series of tests, we added a little bit more spice in the equation. Certainly the job so far was impressive, but add the complexity of guitar signals and things can change. The approach was to try and distant the mix from the original drum set, which still has a lot of reverb.
We added rhythm guitars and solo guitars using different tones from Line 6 Helix. The rhythm guitars have the reverb setting as seen in the picture below, and also were doubled up and panned left and right with a little bit of delay added on one side only to make things a bit more confusing.
The solo guitar is not doubled, but has different amplifier modeller and different settings, and also a more aggressive reverb. The goal was really to make it sound like each instrument came from a completely different room.
For the next step, we decided to start with the same preset used for the drums with our tweaking. That immediately jumped out as sounding artificially dry and unnatural, making the solo guitar in particular sound odd.
This is the nature of the beast of course, as there is a trade off between how much processing you can apply and how natural it will sound. We then started the tweaking, lowering the focus setting a little bit, adjusting the focus bias to a more flat response due to the wider nature of the mix once the guitars were introduced. This got us 80% of the way, with a few tweaks on the other controls allowing us to get closer to a decent sound.
Note that the action in our tests was mainly between the Focus and Localize controls.
Also important to note that we didn't expect this plugin to be able to make the drums sound as if there was no reverb at all, but rather get closer to a sane sounding mix of dry and wet signal as originally intended. The result is actually a bit more noticeable with headphones.
Of course this is subjective so you have to experiment and adapt to your own use cases.
For the tests we’ve utilised a 2017 iMac with 3.6 GHz Intel Core i7, 16GB of RAM and mac OS Mojave, running Logic Pro X 10.4.4.
The CPU load was quite noticeable (around 30% in addition to Logic Pro X), with almost negligible additional RAM consumption even when running multiple instances of the plugin.
The bounced tracks didn’t present any additional artefacts in comparison with the original dry versions without the plugin.
During the testing no crashes or noticeable bugs were observed; the plugin was very responsive and quite easy to use.
The trend of using computational power and artificial intelligence to improve production and fix issues that would otherwise be hard to fix is quite clear. This plugin can certainly rub shoulders with the titans of the industry out there as it clearly seems to do what it promises it will do. The use case tried here is very common especially if you have no control of the original mix, for example in a jam situation or impromptu live recording, or with inexperienced engineers behind the faders.
Have a look at their website for purchasing options - of course variants for all sorts of DAW do exist and be sure to check their other products, which we will cover in future articles:
If you want to purchase the gear we’re used for these tests, here are some Amazon links for RME Babyface pro, Universal Audio Apollo, Rupert Neve DI and Ibanez RG8570z: