IK goes AI

[Burillo]

Hmm? my only real gripe with Amplitube 5 tone wise is certain high frequencies are harsh, i.e. digitally harsh, not solid state or tube harsh. It all makes sense really, how do you oversample live playing? If their take on amp and cab modeling vs Guitar Rig 6 with it's convolution etc. is capable of getting rid of that sometimes in the right (or wrong really) situation artifacts, I'm all in. I probably am anyway, other amp modeling companies sound great but leave something out.

Try this one weird trick:
Put a second mic on the miked speaker, and invert the phase. Move it around and try different volume levels until you get something you like. You can fine tune the characteristics of the phase cancellation through position and volume of the 2nd mic.

The effect of this will smooth out the cab and make it sound more natural, and diminish the "veiled" effect of convolution.

what i like to do is use a ribbon mic pointing into the cone as my base (R121 for example) to get good low end and mids, and mix in an SM57 (or anything dynamic) looking into the coil (or slightly to the side) until i get a nice full balanced sound. that has been my goto into all amp sims from day 1, and it never fails me.

Also, try to use just the SM57 and U87 mic models whenever possible, because the other mics are actually just swapped through (de)convolution.

Source?

[Burillo]

I hope IK uses the Ai Machine Modeling in-house to rework all of the existing amps and distortion pedals in AmpliTube to be ultra-accurate.

They CAN do this if they wanted to. It’s about time for a total overhaul of all of AmpliTube’s gear.

i suspect the tech will have the same limitations Kemper does, e.g. good sound with all knobs at noon, and increasingly digital and fake once you start moving them. i hope IK proves me wrong though

[Funkybot's Evil Twin]

I hope IK uses the Ai Machine Modeling in-house to rework all of the existing amps and distortion pedals in AmpliTube to be ultra-accurate.

They CAN do this if they wanted to. It’s about time for a total overhaul of all of AmpliTube’s gear.

i suspect the tech will have the same limitations Kemper does, e.g. good sound with all knobs at noon, and increasingly digital and fake once you start moving them. i hope IK proves me wrong though

First, I agree that it sounds great in those examples. But if you've used other profiles/profilers, I think that's to be expected. But, I'm glad IK is making this easy. I think Overloud screwed up by not making their profiler available for end-users.

I'm hoping one day someone figures out how to do dynamic captures that can interpolate knob values to create better models. Example:

1. I setup my amp with EQ at noon and volume at 1 - do a capture
2. I setup my amp with EQ at noon and volume at 3 - do a capture, repeat a few times for the usable gain range - the profiler then figures out how the amp response changes at each gain level and interpolates the values creating a more complete profile

Then...

3. Do a capture with volume at noon and Bass EQ (or tone) at minimum
4. Do a capture with volume at noon and Bass EQ at max, repeat for other knobs...the profiler captures the sweep of each EQ or tone knob (maybe you tell it the knob configuration for that channel) and creates a more complete EQ model that matches the amp

Obviously, that would be a lot more complex and take more time to capture, but damn would that be cool.

Then have something that could take all those various data points at various settings and create a model of the amp that includes the amp at different gain and EQ settings for a more complete model.

[Uncle E]

That can be done fairly easily. It won’t be perfect because amp EQ’s react to each other. Would still be great despite that.

An easier way would be to profile clean and dirty settings, then interpolate between the two.

[Funkybot's Evil Twin]

That can be done fairly easily.

I'd hope so too but considering no one's done it yet (to my knowledge) I suspect it's actually pretty hard and would get into "Nebula/Volterra Kernel" territory with major CPU/latency. But I hope I'm wrong. Very wrong.

[Uncle E]

You’re right. Interpolating between filters is easy but making a system that does it live is another matter, particularly between the number of profiles you’re talking about. I could see IK doing it in-house on AmpliTube models.

[noiseboyuk]

https://youtu.be/R6kbE6NgIYc

This latest demo is pretty good, but I can hear a very clear lack of weight in the modelled versions this time. Fortunately it's fairly easily corrected to at least feel a lot closer.

This is the 4 second riff of their 2nd example in the video. The first 2 seconds is real, the second 2 seconds is modelled. Listen to this on good monitors or cans with a solid low end.

https://www.dropbox.com/s/ccsfuwgze2wnl ... x.wav?dl=0

I can clearly hear a drop in low end in the 2nd half - the modelled half.

This second example is the same edit, but this time the modelled version (last 2 seconds) has a +6db shelf at 120hz

https://www.dropbox.com/s/dso8oij69y5w1 ... f.wav?dl=0

They're not identical but do feel much closer now. There's a fairly uniform heft to it.

If there wasn't an error IK made in making the video I think they need to take a look at this in the product itself since every demo example felt lacking in that low end - it was perhaps easiest to hear in that Hendrix-type riff, but it was always there. The hard stuff though - the rest of the sound shaping - sounds stellar, so hopefully this is an easy fix for them to make.

[machinesworking]

This latest demo is pretty good, but I can hear a very clear lack of weight in the modelled versions this time. Fortunately it's fairly easily corrected to at least feel a lot closer.

I think this tech is pretty cool but I really didn't here the loss of bass until I put the original into a two track editor and adjust the AI versions gain to match the real speaker. If you open your links to Dropbox in dropbox even you can clearly see a 1-2db gain on the AI version, which helps to hide the lack of bass... You're right it's not drastic though, and should be able to be tweaked by IK, rather than tweaking the gain on their version to make it sound "better".

I also wish they didn't stick to just one model, hopefully more demos in the future etc.

[noiseboyuk]

I wonder how many will hear that video demo and just think “hmm, not bad but something lacking”, perhaps not having the analytical skills to break it down further. It sounds thinner, not as beefy as the real thing, which psychologically is gonna hit right on the confirmation bias - it’s exactly what sceptical people might expect to hear of a virtual copy of something real. Only in this case objectively it is true. And all they needed to do was lift the bass by 6db!

If I were them I’d take it down immediately and then put up a fixed version. Then fix the product itself! (Though I do wonder if it was just an issue making the promo video).

[jamcat]

You’re right. Interpolating between filters is easy but making a system that does it live is another matter, particularly between the number of profiles you’re talking about. I could see IK doing it in-house on AmpliTube models.

People should be mindful that at the heart of any amp profiler, there is still a more or less standard ampsim model. Machine learning just finds the optimal weights to reconstruct the training data. So with several takes using well-chosen settings on your amp, you could plot a best-fit curve through the training models' weights to precisely reproduce the amp's behaviours one-by-one in a single universal model, just like a standard ampsim model.

This is something that IK would have to do in-house, since only they have access to their modeling algorithms that they're training. They could run all of the amps in their collection through this process and release new definitive models of all of the AmpliTube amps. That is what I was talking about.

[escalona]

It seems to me like they invented a kemper profiler... only 12 years after the kemper profiler...

[jamcat]

It seems to me like they invented a kemper profiler... only 12 years after the kemper profiler...

And without the Kemper Profiler

[jamcat]

Also, try to use just the SM57 and U87 mic models whenever possible, because the other mics are actually just swapped through (de)convolution.

Source?

"IK’s new Volumetric Impulse Response (VIR™) uses 600 IRs per speaker, 1,200 IRs for a 2-speaker cab and 2,400 IRs for a 4-speaker cab"

https://www.ikmultimedia.com/products/a ... /#whatsnew

Each mic can be set straight on, or at a 45° angle, so that means 300 IRs per speaker per angle. There are 30 x 5 non-interpolated impulse positions per speaker. 30 x 5 = 150. That is exactly enough impulses for 2 mics. So to break it down, those 600 IRs come from:

(2 mics x 2 angles x 150 positions) = 600 IRs

You may notice in the video and all the photos from the IR creation process, you only ever see 2 mics shown: the U87 and the SM57. That's because those were the 2 mics that were used.

https://youtu.be/-vac5fHcaY4

The other mics are swapped with deconvolution and convolution to remove the mic impulse then imprint the audio with a new mic impulse. I'm sure you are aware that is also how T-RackS Mic Room works. And you may note that TR Mic Room has the exact same mics available as AmpliTube.

The incontrovertible proof of what I'm telling you can be observed by using TR Mic Room after AmpliTube. Swapping the SM57 or U87 with another mic in TR Mic Room will yield IDENTICAL results as selecting the alternative mic directly in AmpliTube. (Use the SM57 as the source for dynamic mics and the U87 as the source for condenser/ribbon mics.)

Also, AmpliTube uses this same devoncolution/convolution process for swapping speakers. You can hear the shortcomings of convolution by playing the 2x12 Open SL and then swapping the American 12C and Silver Alnico speakers with each other. It should sound the same. But there is a noticible quality degradation when the speakers are swapped.

Another cab you can try it with is the 2x12 Z-Wreck. Swap the blue and gold Brit Alnico speakers with each other, and observe the loss of quality when you do.

Or swap the speakers Brit Alnico B speakers with Brit Anniversary 1 speakers and visa versa in the Brian May Vox cabs and compare the sound. Since they use the same cab, they should sound the same when they have the same speakers. But once again, the swapped speakers are noticeably lower quality.

You can also make these kind of comparisons with the AmpliTube mics and TR Mic Room. Choose a mic that isn't the SM57 or U87 in AmpliTube, then replace it with the U87 or SM57 in TR Mic Room, and compare it to the U87 or SM57 in AmpliTube as the selected mic. You'll notice the Mic Room swapped U87 and SM57 sound the same as their AmpliTube counterparts, but just degraded in quality.

That's convolution at work.

[Uncle E]

It seems to me like they invented a kemper profiler... only 12 years after the kemper profiler...

Better late than never!

[Funkybot's Evil Twin]

Hey, there’s a whole MODI guitars thread in the instrument forum. Discuss it there!

This thread is about the amp profiler thingy that is an audio effect.