Liquid Profiles

Quote from V8guitar

Has it?

Can someone explain machine learning because so far all I've seen is more complex algorithms rather than what I understand as machine learning.

Machine learning to me is taking multiple complex inputs and then apply this to future calculations. With a capture, you capture the input as accurately as possible, so where is the machine learning?

Someone please show me I'm wrong as I'm sure I am....but feels like marketing buzz words...

Display More

I’m no expert, but I found a definition that helps me:

Machine Learning is an AI technique that teaches computers to learn from experience. Machine learning algorithms use computational methods to “learn” information directly from data without relying on a predetermined equation as a model.

Quote from Ruefus

Machine learning algorithms use computational methods to “learn” information directly from data without relying on a predetermined equation as a model.

That is my noob understanding as well.

The Kemper outputs specific sounds and measures the result from the amplifier. The code is looking for specific measured responses.

An AI system would have an input and a resulting output. It would then compare the two and compute formulas to get the result from the input. There are probably several different methods to reduce the data difference into a large set of formulas and data.

The AI code is not looking for a specific response like the Kemper would be.

The Kemper may say at gain .5 @ 1kHz we get this response. The Kemper would then tailor its fixed formulas to the measured values.

The AI model has fixed transformation formulas that it gradually adjusts the coefficents until the calculated output matches the measured output. Its formulas are not based on a guitar amp or anything like it.

Since the Kemper is looking for specific guitar related things, its formulas are modeled after an amplifier. It needs a smaller set of coefficients to make an amp sound. The gain does this and the frequency does this.

An AI model uses generic math transformation formulas. So it needs a lot more information and testing. And the resulting formulas are not efficient since they are designed to handle any possible scenario. There is a lot of number crunching going on. Thus the NAM, for example, is very heavy on CPU usage.

The Kemper could actually be one of the most efficient CPU users out there. Normal modelers are doing all sorts of code to model a specific thing in the amp. So they probably use more CPU than the Kemper. Kemper VST when?

One could argue that an AI approach will be closer to the original. However, the Kemper will allow you to adjust 20 parameters about the amp since it knows its an amp and has measured it. The AI only knows I get data X and output Y. You cant adjust sag, compression, etc. It has no idea what those values are.

(let's still keep this thread clean and talk about the wonderful perspective of Liquid Profiles instead of lawsuits or trolls)

Any news on when this will be available? I am interested in the USB audio part more than in the liquid profiles at the moment.

Quote from rockbandguy

Any news on when this will be available? I am interested in the USB audio part more than in the liquid profiles at the moment.

no, but you will hear it first (sorry, sarcasm)!

It's not available yet.

Just had a look at the new ToneJunkie-Video (and no I don't want to discuss the content of the video).

But some tonestack-options (amp models) are shown:

As I'm not really into Fender or Vox amps, I'm not sure if I got everything right. Please correct me, if I'm wrong.

Quote from RosboneMako

That is my noob understanding as well.

The Kemper outputs specific sounds and measures the result from the amplifier. The code is looking for specific measured responses.

An AI system would have an input and a resulting output. It would then compare the two and compute formulas to get the result from the input. There are probably several different methods to reduce the data difference into a large set of formulas and data.

The AI code is not looking for a specific response like the Kemper would be.

The Kemper may say at gain .5 @ 1kHz we get this response. The Kemper would then tailor its fixed formulas to the measured values.

The AI model has fixed transformation formulas that it gradually adjusts the coefficents until the calculated output matches the measured output. Its formulas are not based on a guitar amp or anything like it.

Since the Kemper is looking for specific guitar related things, its formulas are modeled after an amplifier. It needs a smaller set of coefficients to make an amp sound. The gain does this and the frequency does this.

An AI model uses generic math transformation formulas. So it needs a lot more information and testing. And the resulting formulas are not efficient since they are designed to handle any possible scenario. There is a lot of number crunching going on. Thus the NAM, for example, is very heavy on CPU usage.

The Kemper could actually be one of the most efficient CPU users out there. Normal modelers are doing all sorts of code to model a specific thing in the amp. So they probably use more CPU than the Kemper. Kemper VST when?

One could argue that an AI approach will be closer to the original. However, the Kemper will allow you to adjust 20 parameters about the amp since it knows its an amp and has measured it. The AI only knows I get data X and output Y. You cant adjust sag, compression, etc. It has no idea what those values are.

Display More

This was my point, I couldn't understand the " learn" part. I think I get what you are saying but for me sound more like auto refining....input a sound, analyse, look to match, compare and modify. Nam etc. do this many times, whereas the KPA has a manually selected process.

I'm sure I'm totally wrong/its more complex than that but for me, this is not what I would call true machine learning because the model is pre-determined. Anyway, I don't want to derail the thread....

Quote from MattUlbricht

Fortunately, that last statement is untrue. CK states exactly that the gain knob will behave like a real gain knob. (see HW's Video on YouTube, min. 7:15) But then again, what else should he say, seen from a marketing point of view...

What if the amp doesn't have a gain knob. My Bassman, AC30, DR, and Marshall, don't have a "Gain" knob. They have volume knobs. I like turning up the "Volume" to overdrive those amps. My Bassman sounds amazing cranked up. The amp goes crazy, It sags more, the louder you crank it up. That's what I would want. I don't like fake gain that acts like a distortion pedal. Is that what it's going to do? I don't know how, because of the way all of the components and speakers react to the volume change is part of it. That would be awesome.

Obviously there is almost nothing we know.

We have to wait until the first beta or at least some more announcements from kemper.

What we do now is wild speculations.

Quote from jon9max

What if the amp doesn't have a gain knob. My Bassman, AC30, DR, and Marshall, don't have a "Gain" knob. They have volume knobs. I like turning up the "Volume" to overdrive those amps. My Bassman sounds amazing cranked up. The amp goes crazy, It sags more, the louder you crank it up. That's what I would want. I don't like fake gain that acts like a distortion pedal. Is that what it's going to do? I don't know how, because of the way all of the components and speakers react to the volume change is part of it. That would be awesome.

Are you using an attenuator or are these loud tube amps that you're talking about?

Quote from jon9max

What if the amp doesn't have a gain knob. My Bassman, AC30, DR, and Marshall, don't have a "Gain" knob. They have volume knobs. I like turning up the "Volume" to overdrive those amps. My Bassman sounds amazing cranked up. The amp goes crazy, It sags more, the louder you crank it up. That's what I would want. I don't like fake gain that acts like a distortion pedal. Is that what it's going to do? I don't know how, because of the way all of the components and speakers react to the volume change is part of it. That would be awesome.

Nothing can be all things to all people. I would expect a liquid profile to be closer out of the box. With controls like Definition, tube shape, power sagging, compression (amp), clarity....... you'll probably get there. I preach my love for the back-end control on the Profiler all the time, but I know of no other digital solution that gives you this sort of fine control over a sound.

Tube amp preamps typically have a cascade design where a succession of 12AX7 or similar tubes boost the signal and pass it on to the next stage which boosts it further. The "volume" or "gain" control is just a potentiometer which wastes some proportion of the signal to ground before the wipe of the pot passes the signal on to the next stage.

A "pot" is just like your guitar volume control, a variable resistor with the incoming signal connected to the top of the resistor, the bottom connected to ground and the output taken from the wipe, the position of which can be adjusted with the knob to take an adjustable output level anywhere from full amplitude to zero amplitude. If the "volume" or "gain" pot is up full, then the wipe is connected directly to the input and the pot is more or less out of the circuit in terms of level control - the coupling cap feeding the signal to the gain pot and the resistance of the pot still comprise an RC filter which functions as a one pole high pass filter. That "up full" setting is what you get in the absence of a gain pot at that location - so if there is no additional gain pot the thing is wired to run the amplification through the cascade flat out.

Amps with multiple gain, volume, master volume pots like mesa boogie designs just have more of these pots between the gain stages so that the amount of distortion can be fine tuned at multiple points in the sequence.

Amps with just one gain control typically have it early in the sequence of gain stages (eg directly after the first gain stage which is straight after the input) and, after that gain/volume control, the signal from each half-tube gain stage goes either directly into the next gain stage or on to the tone stack (which typically cuts the signal amplitude quite a bit and therefore another gain stage is needed to boost it back up again).

I'm just excited to have the gain knob be more realistic and hoping it will prevent having to load up additional profiles to lower/increase the gain. If it does that, then I will be happy with the update.

And with all the technical meanderings, it always boils down to does it sound good. Will it inspire me to play my arse off, will the listener enjoy the sound. If all these answers are yes, then we have achieved something great!

My new Album is out, done with mostly the Kemper (Fuchs, Carr and Two Rock profiles). I did sneak my Ceriatone ODS mini and Synergy BEBB module in there.

https://Paulvictor.hearnow.com take a listen

Quote from tron621

And with all the technical meanderings, it always boils down to does it sound good. Will it inspire me to play my arse off, will the listener enjoy the sound. If all these answers are yes, then we have achieved something great!

Theoretically the sound remains the same. Just responds to adjustments more accurately in relation to the original amp. For some this isn't that important but its been requested for many years.

Quote from radiooctave

I have never ever played any of Fractal Audio's gear, but judging on the number (and ranks) of professional musicians who do use it for their work, it is somewhat safe to say that FAS produce very excellent sounding pieces of gear, in particular regarding the effects chain (see, e.g., John Petrucci). So I believe that the sound and most likely also the feel when playing are beyond reasonable doubt, and it comes down to usability, preference of workflow, additional options etc.

A lot of people use it just for the FX without the amp models. I have the axe3 and the floorboard and I use it with my real amps for the routing and FX and "pedalboard". Some of the FX are really good and about everything is covered. It even has a Keely Halo effect, so they keep an eye on whats popular and try to provide it.

I dont particularly like the models in the axe 3 personally.

Quote from Michael_dk

One vendor has already said as much.

Hi, can you point me to whoever made this comment?

My guess is that CK will also give the popular vendors, Bert M, MBritt, TJ etc, an advance opportunity to create new profiles meant to be paired with the initial Liquid stacks, and that Kemper won't be releasing its new firmware until it can offer a nice free rigpack of ready to go Liquid profiles that demonstrates its potential.

Quote from Wreckington

I'm just excited to have the gain knob be more realistic and hoping it will prevent having to load up additional profiles to lower/increase the gain. If it does that, then I will be happy with the update.

I think that's the essence of it. If Liquid Profiles work as advertised, they should eliminate the "each profile is just a snapshot" limitation of the Kemper. The update is not about better sound (which is subjective anyway), and I think "authenticity" is not the best way to describe the benefits either, although you could say the tone stack models will allow profiles to be tweaked more "authentically" (but who cares as long as it sounds good?).

I think the main benefit will be that the decisions and conditions at the moment of profiling will become less limiting. Say I only have access to a certain amp for a short amount of time (because I'm planning to sell it, or because I'm borrowing it). A good amp will have a lot of different sweet spots to be discovered, and at any time, who knows what other settings would sound good too! Until now, there was always the pressure to cover all of them, because those snapshots were all you had left once the original amp was gone. If you failed to profile a particular sweet spot, it would not be possible to exactly recreate or discover it starting from other profiles.

With Liquid Profiles, that worry should be over. You will only need to profile an amp once, not even worrying too much about settings, and that profile will include even those sweet spots you don't know yet. At least in theory, at least for those tone stacks for which a model exists.

So that's the promise. I don't care too much about authenticity for its own sake, but the idea that you can now profile an amp's full potential instead of just specific snapshots sounds pretty cool.

Quote from spotter

1. Fan Twins Rev Nrm	Fender Twin Reverb
2. Fan Twins Rev Brt	Fender Twin Reverb w. Bright Switch
3. Fan D'Lux Rev Nrm	Fender Delux Reverb
4. Fan D'Lux Rev Trem	Fender Delux Reverb w. Bright Switch
5. Fan Vibr. Lux Nrm	Fender Vibro Deluxe
6. Fan Vibr. Lux Brt	Fender Vibro Deluxe w. Bright Switch
7. Fan Vibr. Champion	Fender Vibro Champ
8. Fan BM Nrm	Fender Bassman
9. Voice Ace 15 Topbst	Vox AC 15 w. Top Boost
10. Voice Ace 30 Nrm	Vox AC 30
11. Voice Ace 30 Topbst	Vox AC 30 w. Tob Boost

As I'm not really into Fender or Vox amps, I'm not sure if I got everything right. Please correct me, if I'm wrong.

Some more in the original video and in this demo:

That's a lot of the classics covered already, at just over half of the total of about 40 models CK mentioned.

Quote from Robrecht

Some more in the original video:

12. Roll Land Chor Nrm	Roland Jazz Chorus
13. Mars TM 45 Nrm & Brt	Marshall JTM45 (with & without Bright Switch)
14. Mars Plex Nrm & Brt	Marshall Plexi (with & without Bright Switch)
15. Mars CM 800	Marshall JCM800
16. Fried Man Cln Nrm & Brt	Friedman (with & without Bright Switch)

thanks! I did not see this. Wasn’t looking at the first video, just listening while working

What i missed from TJ's interview were some question's regarding NAM, how it could affect Kemper and it's competitors... (He started a question about ML, AI, but they drifted off).

Would be interesting to hear CK's opinion about that...(maybe not on the question list, which i think is quite common to ask for in advamce)