Equavescent equalizer release

VST, AU, AAX, CLAP, etc. Plugin Virtual Effects Discussion
Post Reply New Topic
RELATED
PRODUCTS

Post

A_SN wrote: It's good to not be the most qualified person in the thread to explain things anymore, so now I can ask questions instead of speculating ;). Do you think what people like about this EQ's sound just has to do with the unusual curves
Yes. And the truncated IRs. I'm fairly sure there's something to that. We'll find out over the next few years. Besides phase, there's literally nothing more to it.

and that you can get the same results with any other EQ that can reproduce those curves
Obviously. That said, reproducing those curves very accurately might well require very short IR lengths. As an EQ designer, it's unlikely to occur to you that that could be something anyone could want. Can you set SplineEQ into ultra lo-fi mode? Like, maybe, 256 samples?
(or what about the phase response?) ?
I reckon it's linear phase most of the time, but anyone with the plugin analyser could take a look.

A simple rule of thumb is this: if you can hear a glaring, obvious difference, then it's the curve shape. If you hear a subtle, delicate difference, relative to other tracks/parts of the mix it's the phase.

And if so do you have any idea what's so great about those types of curves?
Well, they do display a spectacular naïveté.
I could (and might) go into horrible technical depth about the decomposition of meromorphic functions, and why that gives us the shapes we use, and an understanding of the uncertainty principle (sharp in one domain means ringy in the other) is fairly essential, but these just aren't audio shapes. They're engineering shapes.
These have been designed as "design a filter with N dB of attenuation in the stopband, and M dB of ripple". I won't expect anyone to see by eye that there's a discontinuity in the derivative, but there is, and that's going to cause serious ringing. Without resonance, none of our usual EQ shapes will do that.

What's missing is smoothness (in the technical sense). In your SplineEQ (and here I must confess I don't know what constraints you place on your knots), a single spline is guaranteed smoothness because derivatives above three are zero! (Almost everywhere. DC and nyquist are always exceptions).

Well, there's no consideration of smoothness in these designs. These are wacky shapes. They don't transition in a smooth way, they have some unusual ringing discontinuities. And that ringing gets truncated. You see that little ripple as the graph hits flatness? That's a tell-tale sign of a truncated IR. It's only because the edge is so sharp that we can see it at all. Rare to have an example of ripple so visible.

To OP. I'd strongly urge you to avoid further investment in patents on this. DSP as a field is founded on telecoms, and I reckon I can find prior art for this kind of strategy from the 60s. Using PMC like this for audio, from the 70s definitely. I don't have the books to hand, but I'd definitely check the later chapters of Rabiner&Gold and Oppenheim&Schafer. 50years of prior art could leave you with a hideously expensive and completely indefensible patent, if it were to be awarded.
[ DMGAudio ] | [ DMGAudio Blog ] | dave AT dmgaudio DOT com

Post

DaveGamble wrote:
A_SN wrote: It's good to not be the most qualified person in the thread to explain things anymore, so now I can ask questions instead of speculating ;). Do you think what people like about this EQ's sound just has to do with the unusual curves
Yes. And the truncated IRs. I'm fairly sure there's something to that. We'll find out over the next few years. Besides phase, there's literally nothing more to it.
I remember reading in the literature suggestions (or at least speculation) that if an IR has a time-domain discontinuity due to truncation or min-max optimization or whatever (and the discontinuity is far enough from the main peak), said time-domain artifact might be perceptible (at least sub-consciously) even if the frequency domain ripple is negligible.

I have no idea if that's true, but it seems plausible, considering it could affect spatial perception or some such thing. One could compare truncated IRs with and without a short fade-in/fade-out at the truncation points (to smooth out the time-domain artifacts, without much change to the response if the IR is still reasonably long) and see if that makes a difference.

Post

I already did this some years ago.

I wanted to generate the "perfect" low-frequency-highpass-IR for me, which has a lower amount of ringing than usual (especially pre-ringing), and still would have a frequency response, which works well.

I experimented a lot with different fade-ins/outs, further editing of the IRs and different phases (to not loose so much headroom, and still have a "tight" sound).
The results in the end were good ... but somehow I forgot about those IRs, when I later found out, that most of the time a bell- or shelf-filter works better for me in those low-frequency areas.


Regarding the question, why some people seem to like the strange curves of this EQ, I think expectation bias certainly plays a role here.
Usually "normal" curves should sound more natural and unobtrusive. Of course, in some cases those other curves might actually work better ... I doubt, that's the common case though.


I don't think, this mini-ripple in the frequency-area is audible.
Depending on how much ringing is cut off, it might actually influence the "tightness" of the sound though, at least in the low-frequency-area.

It would be cool to have an EQ, where you have a knob for every band, which lets you shorten the ringing (in a "good" way). Would be a cool feature actually.
But no IR-mambo-jumbo please.

Post

Nokenoku wrote: Regarding the question, why some people seem to like the strange curves of this EQ, I think expectation bias certainly plays a role here.
Usually "normal" curves should sound more natural and unobtrusive. Of course, in some cases those other curves might actually work better ... I doubt, that's the common case though.
I agree strongly with you on this. I think also "the cool of the new" plays a big factor. (Edit: A big factor of the 2% that remains after the 98% explained by the unusual frequency response. People were interpreting that sentence as flippant. It needed quantification.)
I don't think, this mini-ripple in the frequency-area is audible.
That's a perfectly testable hypothesis :)
Any volunteers to make some IRs?
Five files, three truncated IRs, two non-truncated, (or two and three) cropped precisely. We all try and classify them, and we run stats on the result.
Last edited by DaveGamble on Thu Feb 27, 2014 1:48 pm, edited 1 time in total.
[ DMGAudio ] | [ DMGAudio Blog ] | dave AT dmgaudio DOT com

Post

DaveGamble wrote:
and that you can get the same results with any other EQ that can reproduce those curves
Obviously. That said, reproducing those curves very accurately might well require very short IR lengths. As an EQ designer, it's unlikely to occur to you that that could be something anyone could want. Can you set SplineEQ into ultra lo-fi mode? Like, maybe, 256 samples?
Yep, if you turn the Resolution knob all the way down then the kernel is the displayed value x1000 (at a 44100 Hz rate), so that'd be 201 samples I think (the kernel, not the whole convolution). And you can see the actual frequency response in the form of the dashed curve. Wait, why would it take very short FIRs?
What's missing is smoothness (in the technical sense). In your SplineEQ (and here I must confess I don't know what constraints you place on your knots), a single spline is guaranteed smoothness because derivatives above three are zero! (Almost everywhere. DC and nyquist are always exceptions).
Well they're just plain cubic Bézier splines in the log-log(ish, there's that linear part below -60 dB) frequency space, but you can bring the dots really close together as to have huge slopes, so the smoothness is really only guaranteed by the windowing. So you can throw any wacky curve at it and the windowing will smooth it out as much as it has to. That's some pretty textbook stuff really, I always find it weird that it seems so many LP EQs work entirely differently (I heard some do things like auto-correlating a minimum phase IR for instance... and then there's this whole trimmed IR thing we're talking about). By the way you can see some rippling in SplineEQ if you lower the Resolution and make a huge brickwall low pass filter with a huge gain.
Well, there's no consideration of smoothness in these designs. These are wacky shapes. They don't transition in a smooth way, they have some unusual ringing discontinuities. And that ringing gets truncated. You see that little ripple as the graph hits flatness? That's a tell-tale sign of a truncated IR. It's only because the edge is so sharp that we can see it at all. Rare to have an example of ripple so visible.
Yeah if you truncate the IR and have something that rises and comes to a dead stop instantly it's gonna ripple (because multiplying by a rectangle in the time domain is like convolving with a sinc function in the frequency domain, that's what we're talking about right?). Which is why I like to assume people do proper windowing. I'm still not sure I understand why they don't by the way, I get that the frequency response gets to be sharper at the cost of more rippling, I just don't understand why anyone would want to go that route, it just doesn't seem to be worth it.
Nokenoku wrote:I experimented a lot with different fade-ins/outs
I merely looked at charts of the frequency responses of many windowing functions (there's a huge Wikipedia article with those) and concluded that a Blackman window would be one of the best choices, and I've stuck with it since then. I think that's the best way to pick those kinds of things.
Developer of Photosounder (a spectral editor/synth), SplineEQ and Spiral

Post

A_SN wrote:And you can see the actual frequency response in the form of the dashed curve. Wait, why would it take very short FIRs?
Very neat. We're using short IRs because we're speculating that a subgroup of people like the sound of truncated IRs.
A_SN wrote: Yeah if you truncate the IR and have something that rises and comes to a dead stop instantly it's gonna ripple (because multiplying by a rectangle in the time domain is like convolving with a sinc function in the frequency domain, that's what we're talking about right?).
That is precisely what we're talking about; though it's mildly unfair, since PMC DOES know about the fact that it's designing a short IR. But only mildly unfair, since PMC doesn't really care about IR ripple; it's trying to match a frequency-domain specification.
A_SN wrote:Which is why I like to assume people do proper windowing. I'm still not sure I understand why they don't by the way, I get that the frequency response gets to be sharper at the cost of more rippling, I just don't understand why anyone would want to go that route, it just doesn't seem to be worth it.
We only guess at other people's motivations. I think this one comes from a first-principles approach to designing an EQ.
[ DMGAudio ] | [ DMGAudio Blog ] | dave AT dmgaudio DOT com

Post

DaveGamble wrote:
kvaca wrote:
bmanic wrote: X = Equavescent low shelf
Y = DMG Audio Equilibrium, tried to null by ear. Could have used a bit more time with the nulling but it's around -30dBFS
what a surprise...I prefered X, so it seems that Equilibrium is definitely not as transparent as I have expected /at least in this particular case/, but to me still X sample sounds like any other common LP EQ I know... :shrug:
Sorry to clarify excessively, but this isn't any kind of demonstration of transparency; this is comparing EQuilibrium being used to match an extremely non-traditional curve shape. I don't mean to imply that you didn't know that; just wanted to add this to clarify your wording for other readers.
Also keep in mind that I tried to match it by ear and put in about 15 minutes of effort. It's definitely possible to match it much closer with more patience.

Snapping an IR of it would probably pretty much null it.

I don't think there's any other "magic" to this thing than weird jargon, extremely magically unreliable (crashing almost constantly) and the strange but somehow nice and strange shapes of the shelves (the asymmetric bells are a nuisance).

Cheers!
bManic
"Wisdom is wisdom, regardless of the idiot who said it." -an idiot

Post

Oh by the way, about the phase of Equavescent. It works exactly like stated on the GUI. There is a phase knob. Set it to 0% and you have fully linear. Set it to 50% and you have something in between and 100% is minimum phase. Verified in VSTplugin Analyzer. Nothing weird about it on the phase front.

Cheers!
bManic
"Wisdom is wisdom, regardless of the idiot who said it." -an idiot

Post

I have opted not to overly engage in the discussion as my original post was aimed at users and has since been overtaken by developers who have time to comment and attack without having time to trial the Beta demo.

There is more to the sound of a crossover than its curve shape. Null Equavescent™ with Logic X Match Eq (match the eq with noise and use Apply at minus 100%) and listen to the nulls. This is a great way to compare a standard fft equalizer with Equavescent™. The null is the subtracted difference.

There are differences to the sound of an eq curve other than the frequency response. I have demonstrated this in Equavescent™. The standard mode is very clear at the crossover and damp mode is very blurred at the crossover. They are opposites. Damp mode crossovers are ugly but are good for cuts that hide crossovers under other signals. This is due to the lack of clarity in damp mode. Listening to the nulls with Logic X Match Eq can also help demonstrate the difference between damp and standard mode. Equavescent™ works well in Logic X and on files that have not already been heavily digitally equalized.

I think it would be fair if those who comment on my topic have given the nulls and the demo Beta product a good try.

Kind regards
lbarratt
www.barrattaudio.com

Post

LBarratt wrote: I think it would be fair if those who comment on my topic have given the nulls and the demo Beta product a good try.
and I think it would be fair if you offer the demo which can be used without continuous crashing...or at least which CAN be installed!!! /on my PC with XP 32bit it fails to install completely/ :wink:

Post

LBarratt wrote:I have opted not to overly engage in the discussion as my original post was aimed at users and has since been overtaken by developers who have time to comment and attack without having time to trial the Beta demo.
We don't need to try the demo to attack your claims, which is what we/I have a problem with.
LBarratt wrote:There is more to the sound of a crossover than its curve shape. There are differences to the sound of an eq curve other than the frequency response.
Yeah? And what would that be (please be specific)? The phase response? What's special about phase in EquavescentTM?
LBarratt wrote:I think it would be fair if those who comment on my topic have given the nulls and the demo Beta product a good try.
No, not when we're criticising you for using senseless jargon, as this has nothing to do with the actual merits of the product.
Developer of Photosounder (a spectral editor/synth), SplineEQ and Spiral

Post

LBarratt wrote:The standard mode is very clear at the crossover and damp mode is very blurred at the crossover.
So what are we talking about here?
Nonlinearities? Modulation?

Post

LBarratt wrote:I have opted not to overly engage in the discussion as my original post was aimed at users and has since been overtaken by developers who have time to comment and attack without having time to trial the Beta demo.
It may look that way, but actually us developers are primarily here because our users have downloaded the demo and asked us to come here, comment, and explain what's going on.

So in actuality, this conversation IS reflective of the users.

Sorry if that's a bit meta, but it's the truth. You pop up on the internet saying mad things about EQ, and people phone me.

Dave.
[ DMGAudio ] | [ DMGAudio Blog ] | dave AT dmgaudio DOT com

Post

This thread has been EXTREMELY useful and informative for me, and I've very much appreciated the contributions of more knowledgeable people.

I haven't tried Equavescent, because it's apparently very unstable, but I've really learned a lot from the discussion hear and I'm grateful for the sharing of expertise.

Post

LBarratt wrote:
I think it would be fair if those who comment on my topic have given the nulls and the demo Beta product a good try.

Kind regards
lbarratt
http://www.barrattaudio.com
This just isn't possible because your plugin crashes _CONSTANTLY_. It is virtually unusable. Like seriously unusable.

Why do you get so defensive? And why are you avoiding all the questions with aggressive "it's special!! listen to me!!" attitude. You have yet to write a single word on what the EQ is doing differently. We've already established that the curves of your EQ are very strange and thus _IMPOSSIBLE_ to match with another EQ. Hence it does not null (though it's possible to get it down to -30dBFS by ear so I'm pretty sure that running an impulse through it would yield a full null). Unfortunately I can't get the damn plugin working in FL Studio which has a great little effect making it extremely easy to create impulse responses.. otherwise I'd have done it already but frankly, even doing it manually in Reaper is such a chore because it feels like walking on thin ice with this plugin.. it constantly crashes. Randomly. Very frustrating. To get those curves in VSTPlugin Analyzer I had to run the damn plugin at least 10 times and be very careful of where and how I click things or it would crash.

Frankly, you have not convinced me at all that there's anything more special about the EQ than the actual curve shapes so if you take that as an attack then so be it.. :roll:
"Wisdom is wisdom, regardless of the idiot who said it." -an idiot

Post Reply

Return to “Effects”