Kemper Profiler MK 2

[Jarick]

If the profiles are being made on the computer I’d think they could figure out batch processing. That one question I have as well. Shooting a bunch of profile input capture and then batch processing would really nice. I do think there’s less need for that though with liquid profiling as you’re typically not needing to shoot dozens of profiles to capture the range of an amp.

 

[Digital Igloo]

People really underestimate the degree of camaraderie that exists in all manner of industries among presumed competitors.

100%. Have known Christoph since 2017, have been good friends with Jason Stillwell (Tone Master Pro's PM) for ~25 years, and I'm ex-Roland/BOSS, so I know and adore a bunch of people there. The Atomic guys are awesome and M@ and Cooper at Fractal are super cool, but I haven't yet met Cliff; would love to buy him a beer one of these days.

I won't lie and say a few companies' executive staffs aren't littered with a few crypto dude bro-like narcissists who step on people and pull shady $#!‡ to get ahead. But this industry talks and we all pretty much know who they are.

 

[BenIfin]

If the profiles are being made on the computer I’d think they could figure out batch processing. That one question I have as well. Shooting a bunch of profile input capture and then batch processing would really nice. I do think there’s less need for that though with liquid profiling as you’re typically not needing to shoot dozens of profiles to capture the range of an amp.

I spent 6+ months with a MK1 Stage using only respected profile maker LP's and they were magnificent at giving you very real, very amp like gain and very amp like eq adjustments over the full range of controls ..... I posted several clips back then demonstrating how much better than non-LP's it is.

LP's with the new 2.0 profiling *should* i.m.h.o be several stages even better ... time will tell.

This is in direct contrast to the total clusterfuck that is the generic gain and generic eq stacks in Tonex / NAM which just sound shitfull once you move them away from their "12.00 " profiled setting ... and that's not even mentioning how all over the shop gain-staging Tonex / NAM is.

With my HX and GP 5, all my NAM Captures are always left stock ... any gaining up or down or eq'ing is done before and after the Capture.

I quietly confident CK may be about to shut-up several people about how what sounds and feels best across the full range of real amp controls as opposed to what null tests the best .... and yes, I know about "parametric NAM captures" ... but these are nothing more than X number of static NAM captures sewn together with a fuck-ton of interpolation ... whereas the LP process combines profiling with component models of real Amp gain stages and real Amp EQ stages.

 

[Orvillain]

these are nothing more than X number of static NAM captures sewn together with a fuck-ton of interpolation

Right, but you understand what that actually means right??

The finer grain your captures, the more accurate to the real thing your entire model is. Let's say you take two captures and you linearly interpolate between them... hello 1980's ... but let's say you take 1024 captures, with a variety of knob positions, and you know and can label each capture with the right meta-data so that your NN system can train the model accurately... and you're interpolation type is of a high quality (IE: something non-linear, that takes more than 2 data points to recreate the interpolated value) then you're simply going to get something that is very accurate to the input.

That is not true with the genetic approach that Kemper and Headrush and others use.

 

[Jarick]

The one reason I think Kemper may really struggle on the capture market is that so many guys running capture devices are running large pedal boards into stereo Tonex One which combined are less than half the cost of a single Player. On the software side there’s no Kemper plugin where there is a Tonex. And then for all in one units the QC is a much more modern device.

So it’s very possible Kemper may win the null test battle (at least for now) but may still lose the current modeling war.

I’m still going to hang on to my Player to see how it all shakes out and test for myself.

 

[Orvillain]

If the new firmware delivers what it promises, I'll be interested to check it out. There's still a fair amount about the Kemper that I don't think really suits my needs, but if this area is improved, it will be very good for them.

 

[quadgear]

That is not true with the genetic approach that Kemper and Headrush and others use.

I don't think Kemper uses genetic algorithms. It's deterministic, right?

 

[2dor]

I don't think Kemper uses genetic algorithms. It's deterministic, right?

Yes - CK actually called it deterministic in the vid above.

 

[Orvillain]

I don't think Kemper uses genetic algorithms. It's deterministic, right?

Regardless, my ultimate point is, the idea that sewing together a bunch of static captures and interpolating between them is somehow inferior to Kemper adding component modelling to their own band-limited capture method... well, that's just nonsense. Parametric captures are the most thorough way to get towards a model of a real thing, outside of component modelling the whole thing.

 

[Chocol8]

I think it is a bit of a stretch to call liquid profiling component modeling. It's applying different tone stack behaviors, but have they said it's actually modeling the tone stack at a component level and also the interaction with the different pre-amp stages before and after at a component level? I would be absolutely shocked if it was component modeling for a number of reasons.

 

[northcountry]

You can find the Kemper patents online. Here is a recent one: https://patents.justia.com/patent/12289586

It's very jargony and drafted to be the broadest possible, as patents usually are. But with a little AI help, the basic concept is:

The core model is “EQ → non-linear wave shaper → EQ.”

Kemper’s patents describe a sound converter made of two linear filters (simple EQ-like blocks) with a simple nonlinearity (a gentle waveshaper) sandwiched between them. The clever bit is how those two filters are derived from your real rig: the Kemper analyzes the amp/cab at a low level and at a high level; it uses the high-level spectrum for the second (post) filter and the ratio of low/high spectra for the first (pre) filter. That combination captures both the amp’s basic tone and how it changes when it’s being pushed.

Take it with a grain of salt because it's an AI-generated description, and there are more details to it, including refinement through liquid profiling, etc. I think the black boxes used by the others take the input signal and train a neural network that keeps adjusting itself until its output matches the real amp as closely as possible.

 

[BenIfin]

Right, but you understand what that actually means right??

The finer grain your captures, the more accurate to the real thing your entire model is. Let's say you take two captures and you linearly interpolate between them... hello 1980's ... but let's say you take 1024 captures, with a variety of knob positions, and you know and can label each capture with the right meta-data so that your NN system can train the model accurately... and you're interpolation type is of a high quality (IE: something non-linear, that takes more than 2 data points to recreate the interpolated value) then you're simply going to get something that is very accurate to the input.

That is not true with the genetic approach that Kemper and Headrush and others use.

I may be totally out of my understanding here, but for Tonex or NAM would not, say, 1024 different knob captures, take at least something like ~90 <-> 120 hours of just individual capturing time before even doing the interpolation ? ... that's like around 8 hours a day for 12 <-> 15 days of just capturing ..... anyone that did that would go insane

So to extrapolate ..... if you have a basic Amp with just 6 x knobs ie: G / B / M / T / P / MV .... to get roughly 1024 capture points across 6 differnet interacting knobs, each knob can only be captured at just over 3 [ 3.2 ] distinct settings .... so that leaves an awful lot of very large "gaps" for the interpolation to be right and accurate in "filling in" ?)

Again, my apologies up front if my math's is fucked up.

 

[BenIfin]

..... and I will add that I am feeling more positive about have grabbed a Player a few weeks ago to try this new 2.0 stuff first hand when it comes out.

 

[Jarick]

I think it is a bit of a stretch to call liquid profiling component modeling. It's applying different tone stack behaviors, but have they said it's actually modeling the tone stack at a component level and also the interaction with the different pre-amp stages before and after at a component level? I would be absolutely shocked if it was component modeling for a number of reasons.

I’ve never heard it called component modeling either. I always thought it was as you described as well, it makes the eq act more like the real amp tone stack at least in frequency settings and makes the gain control more realistic by modeling bright cap behavior. I thought it’s basically all eq but does work very well.

The idea with liquid profiles is to capture an amp dialed in really well at the most gain you’d want, mark the eq and gain knobs from the amp on Kemper so it k owe the starting point, and then it can do the eq thing when you turn the knobs. It’s really effective imo.

 

[BenIfin]

...It’s really effective imo.

Very, very effective .... as long as you use one of the Kemper 30 or so "modelled" Amp Gain and EQ stacks.

Even with the MK1 LP Profiling, a well done LP is a dream to tweak .... it sounds and feels just like adjusting the controls on "a" real tube Amp.

I'm hoping / expecting :) MK2 2.0 + LP's will up this even more.

 

[Orvillain]

I may be totally out of my understanding here, but for Tonex or NAM would not, say, 1024 different knob captures, take at least something like ~90 <-> 120 hours of just individual capturing time before even doing the interpolation ? ... that's like around 8 hours a day for 12 <-> 15 days of just capturing ..... anyone that did that would go insane

So to extrapolate ..... if you have a basic Amp with just 6 x knobs ie: G / B / M / T / P / MV .... to get roughly 1024 capture points across 6 differnet interacting knobs, each knob can only be captured at just over 3 [ 3.2 ] distinct settings .... so that leaves an awful lot of very large "gaps" for the interpolation to be right and accurate in "filling in" ?)

Again, my apologies up front if my math's is fucked up.

Enter, TINA.

 

[[Nathan]]

I may be totally out of my understanding here, but for Tonex or NAM would not, say, 1024 different knob captures, take at least something like ~90 <-> 120 hours of just individual capturing time before even doing the interpolation ? ... that's like around 8 hours a day for 12 <-> 15 days of just capturing ..... anyone that did that would go insane

So to extrapolate ..... if you have a basic Amp with just 6 x knobs ie: G / B / M / T / P / MV .... to get roughly 1024 capture points across 6 differnet interacting knobs, each knob can only be captured at just over 3 [ 3.2 ] distinct settings .... so that leaves an awful lot of very large "gaps" for the interpolation to be right and accurate in "filling in" ?)

Again, my apologies up front if my math's is fucked up.

As mentioned #Tina. There's been robotics attached to hardware units in other applications (access analog, other services like that).
Either way look how "quickly" the big boy companies take to bring out new modelled amps, it would be generous to give them 2 months a pop. So even at the worst end of the spectrum with 12-15 days, its very doable.

I think you'd also be surprised how far some of these software solutions stretch the interpolation. You might look at a popular Guitar Cab suite where you can move the mic around to find your favourite sweet spot, but under the hood there might only be 9 IR captures there. So between automation of amp capturing + interpolation methods it can actually be a pretty fast method to pump out models. Of course you'd expect the more established brands to have more detailed than some of the lesser tier ones, it's upto the reputation of the company to do what they see as best.

I'm sure there's way more involved to all this and there's roadblocks that take time that we can't even imagine cause we dont do it. But just pure capturing alone wouldn't be that bad.

 

[Chocol8]

An automated system could figure out how many captures it needs to make an accurate model. Very few amps would need 1000+, especially since the interpolation doesn’t have to be linear.

That said, it’s all pretty unnecessary unless accuracy trumps tone and usability. The only advantage to tone controls working like they do on a real vintage circuit is because we are used to that. Those tone stacks actually suck and are very limited.

 

[2dor]

An automated system could figure out how many captures it needs to make an accurate model. Very few amps would need 1000+, especially since the interpolation doesn’t have to be linear.

That said, it’s all pretty unnecessary unless accuracy trumps tone and usability. The only advantage to tone controls working like they do on a real vintage circuit is because we are used to that. Those tone stacks actually suck and are very limited.

ChatGPT suggested a total of 125 reamps for an amp with Master, Presence, Bass, Mid, Treble & Gain knobs.

 

[laxu]

I may be totally out of my understanding here, but for Tonex or NAM would not, say, 1024 different knob captures, take at least something like ~90 <-> 120 hours of just individual capturing time before even doing the interpolation ? ... that's like around 8 hours a day for 12 <-> 15 days of just capturing ..... anyone that did that would go insane

So to extrapolate ..... if you have a basic Amp with just 6 x knobs ie: G / B / M / T / P / MV .... to get roughly 1024 capture points across 6 differnet interacting knobs, each knob can only be captured at just over 3 [ 3.2 ] distinct settings .... so that leaves an awful lot of very large "gaps" for the interpolation to be right and accurate in "filling in" ?)

Again, my apologies up front if my math's is fucked up.

Neural DSP uses some randomization to pick the ranges between the extremes for each control, and this seems to work well to train the neural network without having e.g 10^number of knobs permutations.

Similarly their Tina system is for automating this so someone doesn't need to sit there turning knobs. But still, there's really very little evidence that this is working for them when the QC hasn't had new amp models in ages that are not just plugin conversions.