What's The Difference Between A "Tone Capture" And An "Amp Sim" ?

[jay mitchell]

Regardless of technicality, no one refers to tonex, kemper or nam profiles/captures as “models”.

They are presets that were built using an automated process. At their core, those presets contain models, i.e., algorithms that model various behaviors of amplifiers. Whether or not you characterize them as such, the devices you list are every bit as much modelers as are Axe-Fxs, Helixes, etc., etc. Saying they are not modelers is just allowing your susceptibility to marketing-speak to override reality.

 

[Dom Mcsweeney]

TLDR:
Black Box modelling vs. White Box modelling.

Different ways to do something fairly similar in the end, albeit with different advantages and *disadvantages.

*Those can be mitigated when speaking of parametric modelling like the latest releases by ML Sound Labs or Artera DSP, or Neural DSP, or other companies leveraging parametric black box modelling.

 

[Chocol8]

In my mind a digital process simulating the behavior of an amp is an amp sim. That can be done via component modeling (simulating the circuit) or with a capture/profile (simulating the output signal).

 

[DrewJD82]

I think until you break these down to the actual mechanism in which they perform their function, you can roll around in semantics all day long and never make any progress.

What’s a sampler do? Records a piece of audio and plays it back.

What’s a profiler/capture device do? Records a piece of audio and plays it back.

What’s an amp modeler do? Provides a digital recreation of an amp.

What’s an amp profiler do? Provides a digital recreation of an amp.

 

[Chocol8]

What’s a profiler/capture device do? Records a piece of audio and plays it back.

Not really. A profiler/capture is basically measuring a non-linear transfer function and then modeling that. It more like an advanced curve fit algorithm than a recording.

An IR on the other hand is more like a recording.

 

[TSJMajesty]

I think until you break these down to the actual mechanism in which they perform their function, you can roll around in semantics all day long and never make any progress.

You must be at work.

 

[[Nathan]]

They are presets that were built using an automated process. At their core, those presets contain models, i.e., algorithms that model various behaviors of amplifiers. Whether or not you characterize them as such, the devices you list are every bit as much modelers as are Axe-Fxs, Helixes, etc., etc. Saying they are not modelers is just allowing your susceptibility to marketing-speak to override reality.

Sounds like you’re describing a kemper. Tonex and NAM don’t have underlying models they are built and trained based on a machine learning process.

Thanks for the lecture though

 

[Richard_G]

Are you sure this is not applicable to Tonex and NAM as well, @[Nathan]? Source?

 

[FuzzyAce]

 

[Mike Oliphant]

Are you sure this is not applicable to Tonex and NAM as well, @[Nathan]? Source?

My understanding is that Kemper-style profiling involves taking a captured signal and matching it to one of a set of pre-fabbed distortion models. That plus some EQ matching.

That isn't what Tonex and NAM do. They have no pre-configured notion of how amplifiers work. It is purely machine learning based on an input/output signal pair.

 

[Deleted member 472]

 

[[Nathan]]

Are you sure this is not applicable to Tonex and NAM as well, @[Nathan]? Source?

100%, completely different

 

[paisleywookiee]

 

[Richard_G]

Are you sure this is not applicable to Tonex and NAM as well, @[Nathan]? Source?

100%, completely different

Source?

 

[jay mitchell]

Sounds like you’re describing a kemper. Tonex and NAM don’t have underlying models they are built and trained based on a machine learning process.

The "machine learning process" makes a series of guesses using underlying models and then refines those guesses to end up with - wait for it - a model. By definition, any algorithm that simulates a physical process is - you won't believe the answer - a model. Calling it something else based on the superficial process used to assemble and refine it is an exercise in self-serving sophistry, if you'll pardon my alliteration.

 

[DrewJD82]

Not really. A profiler/capture is basically measuring a non-linear transfer function and then modeling that. It more like an advanced curve fit algorithm than a recording.

An IR on the other hand is more like a recording.

break these down to the actual mechanism in which they perform their function

 

[[Nathan]]

Source?

I dont have the time to compile a bunch of scattered information on how the kemper works underneath the hood, it works exactly how Jay and Mike described. If you look yourself and find contradicting evidence to show that NAM and ToneX work this way then please share it (hint: they dont).

The "machine learning process" makes a series of guesses using underlying models

Incorrect. There is no "underlying models" in the NAM/ToneX process.
Context is king and if you're leaning on the technicality that a wavenet "model" and Fractal/Helix amp sim "model" are both "models" then I guess we can throw in there that NAM, Helix Amps, Cindy Crawford & Toy Cars are all models.

Kemper takes audio and tries to match the internal "models"
NAM/ToneX take 2 audio pairs and create a Neural Network "model" from scratch, there is no underlying models.

 

[Dom Mcsweeney]

Once again.

They are all by definition various 'black box' models, or, otherwise, they are 'white box' models.

I would class Kemper in the same black box group as Tonex, NAM, NDSP, et al - albeit different methods to the same ends.

 

[Deleted member 1726]

 

[Deleted member 472]

Once again.

They are all by definition various 'black box' models, or, otherwise, they are 'white box' models.

I would class Kemper in the same black box group as Tonex, NAM, NDSP, et al - albeit different methods to the same ends.

This.