IK and the Amplitube line really sucks in this regard..

[Bombadil]

The rooms, cabs are all different, better, according to IK, with their VIR setup.
The best advice has been given. Make stems of the tracks. Doesn't help at this point, I know.

[Scotty]

The 32bit wrapping can be glitchy but it might work well enough for someone to capture the stems. That is if they can get the presets to load as per my previous post.

Problem is, when the standalone vsts for their collections went to A4, the standalones stopped being updated so all are stuck in 32 bit. It's more trouble than it's worth to use a wrapper as things get glitchy.

[Funkybot's Evil Twin]

I can't see a modeler like the Kempler for instance giving that different a result if using an older version of it. (If such a thing is available)

I'm going to be a bit pedantic here, so apologies in advance, but the Kemper isn't a modeler, it's a profiler like TONEX.

Each Amplitube.dll is a new .dll. Each has offered more features, new amps, improvements to various aspects of the modeling, different cabs, etc. I know it's a pain in the ass if you didn't print stems 7 years ago, but I don't think that should hold back future updates to the sound. We all want "more realistic" and honestly, AT5s best models still could benefit greatly in that regard. If they continue to iterate and make improvements, I think they should.

For me: I wish they'd start fresh with AT6. Focus on a small core number of amp/fx models, and do the absolute best job they can with those. Example: 1) Fender Tweed (Bassman or Tweed Deluxe), 2) a Fender Blackface (Deluxe Reverb), 3) Vox AC-30, 4) Marshall Plexi, 5) Marshall JCM-800, 6) something for the metal heads (Mesa, Diesel, Engl, whatever). Then just some core stomp boxes, but again, killer models of them: DS-1, Rat, Muff, Fuzz Face, Screamer, CE-1 or CE-2, BF-2, Vox wah, spring reverb, existing X-Gear pedals to round out that, then add some T-Racks effects for the Rack and done! AT5 is bloated, has a worse interface than AT4, uses too much RAM and is slow to load, but worst of all suffers from quality that varies widely from package to package. I'd rather they start fresh. Maybe keep AT5 up to date for OS changes, but start fresh with AT6 and beyond.

[jamcat]

For me: I wish they'd start fresh with AT6. Focus on a small core number of amp/fx models, and do the absolute best job they can with those. Example: 1) Fender Tweed (Bassman or Tweed Deluxe), 2) a Fender Blackface (Deluxe Reverb), 3) Vox AC-30, 4) Marshall Plexi, 5) Marshall JCM-800, 6) something for the metal heads (Mesa, Diesel, Engl, whatever). Then just some core stomp boxes, but again, killer models of them: DS-1, Rat, Muff, Fuzz Face, Screamer, CE-1 or CE-2, BF-2, Vox wah, spring reverb, existing X-Gear pedals to round out that, then add some T-Racks effects for the Rack and done! AT5 is bloated, has a worse interface than AT4, uses too much RAM and is slow to load, but worst of all suffers from quality that varies widely from package to package. I'd rather they start fresh. Maybe keep AT5 up to date for OS changes, but start fresh with AT6 and beyond.

This.

But, people have been saying this since at least AT4. It seems IK would rather just keep adding to the hoarder house that AmpliTube has become.

I’m looking for a way out for this reason.

[Funkybot's Evil Twin]

This.

But, people have been saying this since at least AT4. It seems IK would rather just keep adding to the hoarder house that AmpliTube has become.

I’m looking for a way out for this reason.

I think it's unlikely to happen. But...there is one big difference: now they have TONEX. So in a way, they're competing with themselves. If AT6 is just "everything in AT5, with tiny improvements and a LOT more" then I won't bother. I'll stick with TONEX.

If AT6 is, "hey, we're starting fresh with crazy high quality models" (like I outlined above), then I'd absolutely be in.

[jamcat]

Machine learning is the next big thing in modeling. If IK doesn’t reinvent AmpliTube and T-RackS with next-gen fully modeled gear ML models, then someone else will fill the void.

It’s already happening with complete stomp and preamp neural network emulations from Fazertone and WAVDSP. These are not just static snapshots.

If IK can’t or won’t evolve, they will die.

[Funkybot's Evil Twin]

Machine learning is the next big thing in modeling. If IK doesn’t reinvent AmpliTube and T-RackS with next-gen fully modeled gear ML models, then someone else will fill the void.

It’s already happening with complete stomp and preamp neural network emulations from Fazertone and WAVDSP. These are not just static snapshots.

If IK can’t or won’t evolve, they will die.

How are those Fazertone pedals? On par with TONEX just with full range of settings?

[Bombadil]

In my main templates, the only remaining Amplitube patch is a Leslie model, which I find quite nice. I'm slowly rolling new Tonex patches into Amplitube. Have a Twin and a Plexi, next probably a VOX 30, using Sunset Studio live room and spring plate. I can play them for hours. I'd also say there's no noticeable latency with these as inserts on the channel. Pretty amazing sound, really.
About what I was hoping for with my PODxt….

[jamcat]

Machine learning is the next big thing in modeling. If IK doesn’t reinvent AmpliTube and T-RackS with next-gen fully modeled gear ML models, then someone else will fill the void.

It’s already happening with complete stomp and preamp neural network emulations from Fazertone and WAVDSP. These are not just static snapshots.

If IK can’t or won’t evolve, they will die.

How are those Fazertone pedals? On par with TONEX just with full range of settings?

At least. There is a free Klon Centaur. I have no need for the Centaur in TONEX now. And the free Dyna Comp… wow!

Try them for yourself!

[CrystalWizard]

nm

[Immanencer]

My prediction is that the Amplitube line and TONEX line will be technically unified starting in A6. (I won't speculate whether they'll keep the product sales separate for marketing reasons...) Right now they are compatible (side-by-side) but there's only so far an existing engine can be advanced before a whole new version makes sense technically. I have no inside contacts at IK, but it seems to be less to do with 32 vs 64-bit and more due to the trend from A3 to A4 to A5 (and beyond) of replacing component modeling with machine learning/profiling. This field (along with other AI applications) has really taken off in the last year or so, as most are now aware. As somebody who grew up with physical pedals and amps, I've found that each version has becoming increasingly close to sounding like the authentic, across the entire range of included products. The jump between A4 and A5 was enormous! TONEX is just taking that behind-the-scenes process (used in A5) one step further. Comparing the same pedals or amps between the versions (using the same cabs, of course) highlights how much their profiling has improved. (A3 stuff was so far off from the original because it wasn't yet based on machine-learned profiles but rather fiddling with component modeling until it sounded "in the ballpark".)

IK apparently does further research between major releases, and these incremental advances find their way into intermediary packages. Eg., the original Mesa-Boogie add-on in 2015 was a huge step forward in realism, but this occurred between A3 and A4. Likewise with the Satriani package between A4 and A5 (just compare its CPU use with other models - a good indication of a technological variance from the standard models), and TONEX between A5 and A6.

It's basically impossible to make old presets work between entirely different architectures, but there are no problems with having A3+A4+A5 side-by-side on your system or within the same project. (It works fine, but personally anything before A5 sounds terrible to me now in comparison.)

[Scotty]

We would need to be a Phd AI specialist to really know what IK is up against in terms of AI modelling cabinets/volumetric spaces and various microphones within those spaces.

Impulse modelling and variations of it (Volterra Kernels) is very impressive and effective. For an AI equivalent, I suspect the compute power needed would be enormous. Research into how to put into code the machine learning training models is likely more involved than modelling static captures of a single setting of a specific signal chain.

It would make sense that TONEX technology would be able to make discrete captures of distortion stomps at some point as that is at least related to an amplifier circuit. IK even suggests capturing the amp with the distortion stomps engaged if you want to capture that specific signal chain at those specific settings. Acoustica Audio plays in this arena and they do allow some flexibilty with the settings.

If AI modelling of speakers/cabinets, with various mics within rooms were possible, there is no guarantee that the results would be better than the VIR technology that is in place now. VIR is an amazing technology. Tonex is an amazing achievement. The marriage of the two at that incredible price point is at least impressive if not revolutionary.

Anyone prognosticating IK's demise if they don't implement VIR in an AI form immediately is getting ahead of themselves.

Machine learning is the next big thing in modeling. If IK doesn’t reinvent AmpliTube and T-RackS with next-gen fully modeled gear ML models, then someone else will fill the void.

It’s already happening with complete stomp and preamp neural network emulations from
If IK can’t or won’t evolve, they will die.

[jamcat]

Impulse modelling and variations of it (Volterra Kernels) is very impressive and effective. For an AI equivalent, I suspect the compute power needed would be enormous. Research into how to put into code the machine learning training models is likely more involved than modelling static captures of a single setting of a specific signal chain.

Impulses don't sound very impressive. Convolution is linear. And it's lossy, because of the constant of integration in the antiderivative calculus at the heart of convolution. It's also quite computationally expensive, again because of the integral calculus involved.

Machine learning packages for Python already exist. This is what everyone who is doing ML is using. I'd say a cab model built on an ML neural network inference engine is going to be both smaller and computationally less expensive than an IR of the same. And the ML model will be dynamic and life-like, while the IR is static, grainy and veiled sounding.

It would make sense that TONEX technology would be able to make discreet captures of distortion stomps at some point as that is at least related to an amplifier circuit. IK even suggests capturing the amp with the distortion stomps engaged if you want to capture that specific signal chain at those specific settings. Acoustica Audio plays in this arena and they do allow some flexibilty with the settings.

If AI modelling of speakers/cabinets, with various mics within rooms were possible, there is no guarantee that the results would be better than the VIR technology that is in place now.

There are already discreet captures of distortion stomps in TONEX. And current Machine Modeling is capable of modeling a cab, too, of course. TONEX can even automatically separate the amp and cab components from a profile. And if IK can build a matrix of IR responses covering a 3-dimentional field in front of an amp, then they can create the same matrix using ML captures. It's just a matter of the time (and therefore cost) involved. But a ML-powered VIR would objectively sound better than the current IR-powered VIR.

Once again, this is something you can test out for yourself, quite easily. In either TONEX or AmpliTube, simply play a TONEX amp, and then bypass the cab component of the Tone Model and enable the VIR version of the same cab. Set up 2 instances in your DAW so you can switch back and forth quickly and easily. The inferiority of convolution will become quite apparent.

Anyone prognosticating IK's demise if they don't implement VIR in an AI form immediately doesn't know jack and that is a bad thing if you plan on plugging your guitar into something.

Of course that's not what I said. I said that other developers surely will be developing full amp emulations with ML, and if IK doesn't clean out the mess that is currently AmpliTube and do the same, they will become obsolete. The problem with IK is that as long as management thinks there's a buck to be made on the old, mediocre gear models, they will remain in AmpliTube, even though they're watering down the brand.

[Scotty]

I won’t argue over what you feel might happen. If we are talking opinion that’s fine. I would welcome someone with real computer science and machine learning qualifications to weigh in on the topic. Sometimes people talking sh*t doesn’t cut it.

I am not sure where you derive your negative view of IK's management. We’ve had a steady stream of innovations from IK in software and in hardware but again that is just opinion.
I do take stock in the numerous industry awards that have followed the company and they do seem to me to be capable of innovation in the numerous spaces in which they compete.

I do believe that VIR could be improved and possibly with AI down the road but I highly doubt management is resting on their laurels and greedily holding back innovation for the profit motive as you confidently assert.

Knowing python is a preferred language for ML doesn’t make the code or the compute power appear. Maybe we’ll see GPUs step into the role as AI coprocessors, perhaps Intel and AMD will add more AI features to their processors as Apple has done with their M series.

Since we are both comfortable talking out of our arses here is some pure speculation on my part…

IK will not see its demise. AI development takes time, we'll see several iterations of Tonex and may see AI enhanced versions of VIR. In the meantime, countless live performances and studio tracks will be made despite the crippling deficiencies you claim to hear with VIR technology. I must confess I am somewhat insensitive to these negative attributes although I acknowledge they are limited for the technical reasons you described. As discussed previously and elsewhere, component modelling isn’t dead for reasons of flexibility alone. I see a hybrid future with AI making inroads and positively correlated to available computational power.

It is a great time to be living through this. Cool to chat with you even if I’m frequently at odds with you conclusions.

Impulse modelling and variations of it (Volterra Kernels) is very impressive and effective. For an AI equivalent, I suspect the compute power needed would be enormous. Research into how to put into code the machine learning training models is likely more involved than modelling static captures of a single setting of a specific signal chain.

Impulses don't sound very impressive. Convolution is linear. And it's lossy, because of the constant of integration in the antiderivative calculus at the heart of convolution. It's also quite computationally expensive, again because of the integral calculus involved.

Machine learning packages for Python already exist. This is what everyone who is doing ML is using. I'd say a cab model built on an ML neural network inference engine is going to be both smaller and computationally less expensive than an IR of the same. And the ML model will be dynamic and life-like, while the IR is static, grainy and veiled sounding.

It would make sense that TONEX technology would be able to make discreet captures of distortion stomps at some point as that is at least related to an amplifier circuit. IK even suggests capturing the amp with the distortion stomps engaged if you want to capture that specific signal chain at those specific settings. Acoustica Audio plays in this arena and they do allow some flexibilty with the settings.

If AI modelling of speakers/cabinets, with various mics within rooms were possible, there is no guarantee that the results would be better than the VIR technology that is in place now.

There are already discreet captures of distortion stomps in TONEX. And current Machine Modeling is capable of modeling a cab, too, of course. TONEX can even automatically separate the amp and cab components from a profile. And if IK can build a matrix of IR responses covering a 3-dimentional field in front of an amp, then they can create the same matrix using ML captures. It's just a matter of the time (and therefore cost) involved. But a ML-powered VIR would objectively sound better than the current IR-powered VIR.

Once again, this is something you can test out for yourself, quite easily. In either TONEX or AmpliTube, simply play a TONEX amp, and then bypass the cab component of the Tone Model and enable the VIR version of the same cab. Set up 2 instances in your DAW so you can switch back and forth quickly and easily. The inferiority of convolution will become quite apparent.

Anyone prognosticating IK's demise if they don't implement VIR in an AI form immediately doesn't know jack and that is a bad thing if you plan on plugging your guitar into something.

Of course that's not what I said. I said that other developers surely will be developing full amp emulations with ML, and if IK doesn't clean out the mess that is currently AmpliTube and do the same, they will become obsolete. The problem with IK is that as long as management thinks there's a buck to be made on the old, mediocre gear models, they will remain in AmpliTube, even though they're watering down the brand.

[jamcat]

I won’t argue over what you feel might happen. That’s fine. I would welcome someone with real computer science and machine learning qualifications to weigh in on the topic. Sometimes people talking sh*t doesn’t cut it.

I have a computer science degree and I’ve implemented Apriori, Perceptron, Naïve Bayes, and Decision Tree classification algorithms with sklearn in Python.

And I have also solved convolution integrals by hand.