So, is it a dishonor that whoever created a Dyn EQ that many audiophiles consider the best (Kirchhoff) doesn't understand anything about the most elementary functioning of the best Passive EQ?

 

And that sht distorts man. I never met a pultec emu with that level of harmonic distortion (the last one that i met that i can remember distorted a lot was Noiseash's), and i do know that good sounding HW pultecs doesn't distort a lot. If you dare to use this thing on your projects, lower the level of harmonics to around 30%, please.

 

I agree about that Pultec emu, but imho it can be useful in certain situations, like a LoFi buss which can be creative because it sounds very weird I have to say. But it's a creative approach, not pure mixing. I have a chain that I used in a couple of project lately with AA Taupe plus that DV Pultec, you can emulate a recording from the 50s with those 2 guys

 

The Pultec "emus" use a new tech they developed for the captures, they explain it in the website. The thing about it is that DV plugins also sucked in 1.0, so I expect the Pultecs to get better in following updates.

 

No, TDR Nova is the best but again im no audiophile. I don't think the devs don't understand Pultec, it's the machine learning tool and method they've used to get the result for Pultec isn't accurate. The AI have 0 clue how Pultec works, internally. The method might have worked with straight forward active EQs but passive EQs, you get nothing meaningful by only capturing the input vs output especially when it's famous for bands interaction. Same goes for the parallel EQ like the 8200. That should explain why no ML parallel EQ yet.

 

What do you mean?

 

Then, Kirchoff is a good clone, not the best, and pseudo-Pultecs, presented as good stuff by the dev, are bad clones because they were unable to capture the models. Right!

 

Watch please the video presented by Stevie Dude. My question is based in his argument and in the demonstration of his video, here: https://audiosex.pro/threads/tbtech-deep-vintage-emulations.78490/#post-819528

 

The Motown/Hitsville EQ is a very famous unit that has no captures yet, and its a parallel eq. I do believe you're totally right because those band interactions are not that simple. No Massive Passive, not even for color tone.

 

Acustica Rust?

 

Love these debates about music software. I missed them a lot since in 2013, as "faloprati", I stopped visiting regularly this forum for professional reasons. Please consider that divergences in ideas are healthy in my understanding. If you consider otherwise, I still accept your disagreements with my ideas. It was already like this between 2008 and 2013. Happy New Year!

 

It's Machine Learning? I though it was Volterra Kernels.

 

Kirchoff's Pultec curves are not ML result. It's fully algo and that's why I said it's not like the devs don't know. They behave almost alike Pultec without needing the second band, the "shape" knob is a good workflow addition, I like the idea. The curve result is what one would've expect from a Pultec. It's just not the sound for my taste, it's not big enough for some reason, just more bass. While the tube amplification stage is a main part of the Pultec "sound" I think it missed something else. No scientific proof argument for that right now but my guess is it missed the phase distortion effect from Pultec's minimum phase shelves interaction that do something to the low end transients. Didn't get it with Kirchoff but a similar plugin that has same feature, DMG Equilibirum (on Pultec curve) does it perfectly, at least for how I want it to sound. My go-to for clean (no saturation) Pultec is the PA Bettermaker EQ232D plugin, another one that I think done right. Additional bands come in handy sometimes and I just like the GUI. My fav all-out Pultec emu for years is Overloud EQP. It's the definition of the "big" "huge" I love from a Pultec plugin but it distorts easily. Simple clip gaining will solve it.

 

This is something i was thinking earlier: convolution/impulse response is an technique that allows to retain filter caracteristics in a very precise way. And machine learning doesnt seem to have the same accuracy as this 'ancient' technology. They can handle the amp modeling really well, but for filter interactions they just seem to dumb. While convolution, except for Acustica Audio's Dyanmic Convolution/Volterra Kernels tech, couldn't generate harmonics or modulation on any scenario by deafult.

 

Is it possible that the machine hears something Plugin Dr. doesn't?

 

I don't know for sure, that's beyond my knowledge. PD is not really comprehensive tool as it struggles with program dependent processing for how the test material it used. Parallel processing are totally beyond its capability for now where it can only tell the result of the processing while guessing what it is about, not that accurate. Add program dependability to the processing, PD will almost always missed it. Parallel multiband, transient response across the whole frequency spectrum are all beyond its capability too.

while I think the AI tool maybe capable of more, from what I know those machine learning can only separate the result of its learning into two, EQ/Filter or Waveshaping (which what DSP are mostly all about) and slowly trying to decode the settings, combinations and stacks of those two processing until it find one that can match the result of the tested hardware. It takes the input of the audio then try to match its output with the result. That is all it can do basically. PD technically has more tests going on but it measures one processing at a time while the AI is doing everything in one go. That's how I understand it.

Hardware gear however has its own complicated internal chain and stages of processing inside of it, think of multiband, splitting, parallel and serial. Ain't no way to know if you just measure the output right ? The next ideal way to do it, which is component matching, or measuring each stage would put us back to the usual way of the circuit emulation pre machine learning days. Until some genius come up with a way, this is all we get, for now.

 

It's kinda useless atleast to me if it doesn't come with the saturation and sound stage

 

Yes, that's for sure. So, in the end, they can mimic the behaviour on the output, but the machine can't decode the signal path of that gear just by passing through. Let's be realistic: no way an AI can do the signal path of one of those alien HW eqs like SPL PQ. I think even on simpler circuitry will lead us to:

But machine learning has a potential to make devs go the usual way with a new level of accuracy. We can expect more A/B tests like the ones on TBTech website, HW vs Plugin if they implement old DSP tricks to those captures. Far from what it's meant to be, but with good effort i believe good stuff could appear.

Right now i'm just praying for the raise of the Giancarlo of Machine Learning

 

from what i understand Giancarlo and the gang are well into the machine learning side of things. He has teased it for a while now, stating they are close to their desired outcome. I thinkkkkk he said they will integrate it into nebula, and that it is indeed some sort of hybrid AI, DSP, and voltera.

 

Interim correspondence table update:

Brit73 - Graphically Neve 1073
Brit84 - Graphically Neve 1084
US Rack - Presumably a graphical API
Thick Pre - Graphical suggestion of Chandler Germanium Preamp
TransformerX - Vague suggestion of Rupert Neve Designs MBT's Color Comp section
Green AD - Chromatically Burl B2 Bomber ADC
Nylon - Vague suggestion of Rupert Neve Designs MBT's Super Silk section
Tube Shelf - Graphically a Pultecmania design
Tube Bell - Graphically a Pultecmania design
Tube Filter - Graphically a Pultecmania design
Tape S - Studer's putative suggestion via a simple "S"
Tape M - Charade to suggest an "M" coded german tape machine