Tonex -vs- NAM ..... the Tonex "Aliasing" Low Pass Filter is at ~ 14kHz

[EOengineer]

We’re not writing scientific papers we’re shooting the shit.

I’ve peer reviewed this finding and concur.

 

[Orvillain]

Latest version of ToneX. -88dB threshold. No cab enabled. Same capture as the previous VH4 one.

 

[Orvillain]

Hammerstein plot of the same VH4 capture, with no cab.

 

[Orvillain]

Linear analysis. random noise going into ToneX. Again, no cab.

 

[Orvillain]

Linear single tones going into ToneX:

 

[DLC86]

In related news, IK appears to finally have removed the false marketing claim of "... no aliasing whatsoever" (see the second bullet point on the right in the screenshots below):


March 2025:

View attachment 40631

November 2024:

View attachment 40632

If they see this thread maybe they'll also remove "ultra-precise reproduction of high-frequency content"

 

[Orvillain]

One last random noise going into ToneX, but zoomed out:

Seems to be about a 5dB roll off in the high frequencies.

Here's a different capture, it is called raw classic, and the cab is disabled. It is a clean character capture.

So yes... for this capture at least, the amp block does seem to have a bit of a roll off that kicks in around 5000hz. Now whether that filtering is to do with anti-aliasing.... I kind of doubt to be honest. If it is for that reason, then it is completely fucking useless.

I'm more of the mind that this is an aspect of their capturing technology, as @northern_fox pointed out previously.

It remains to be seen whether this roll-off always happens. But over time my ears have told me that ToneX is darker than the real amp I've captured with it.

 

[Orvillain]

Pink is still the clean ToneX capture.

Red is Amplitube 5's 65 Deluxe Reverb.

I tried a few other AT5 models, and they all look different. So even within Amplitube, this sort of thing is not consistent.

 

[DLC86]

@Orvillain I think without a comparison with the reference gear you can't really tell if the high-frequencies are attenuated or not... And as said it seems to vary based on the capture.

I also think it has nothing to do with anti-aliasing cuz that would be pointless, more likely a limit of the neural network they're using.

Anyone wants to talk a bit about the shitty SRC they use which adds additional aliasing?

 

[Orvillain]

Versus Helix Native's US Double Nrm.

 

[Orvillain]

I think without a comparison with the reference gear you can't really tell if the high-frequencies are attenuated or not... And as said it seems to vary based on the capture.

You can compare to other amp models and captures within the ToneX and Amplitube eco-system to derive a pattern. So far, I'm not able to see one. Which weakens the opening conclusion.

Furthermore, the harmonic analysis sweeps show a much more detailed picture across the dynamic and frequency ranges, and there is no sign of a fixed lowpass filter causing a fixed frequency roll-off.

So the claim that there is a 5k roll off in order to tackle anti-aliasing, I think we can safely put that to bed.

Ultimately, I cannot reproduce what Ben has reported here.

 

[Orvillain]

The one pattern I can see is that with my amp captures that were "head only" there is a commonality around 14kHz - all of my captures have some kind of roll off, and it isn't a nice pretty curve either:

And I will go on record as saying that a real amp does not have this kind of high-frequency drop.

 

[DLC86]

You can compare to other amp models and captures within the ToneX and Amplitube eco-system to derive a pattern. So far, I'm not able to see one. Which weakens the opening conclusion.

Furthermore, the harmonic analysis sweeps show a much more detailed picture across the dynamic and frequency ranges, and there is no sign of a fixed lowpass filter causing a fixed frequency roll-off.

So the claim that there is a 5k roll off in order to tackle anti-aliasing, I think we can safely put that to bed.

Ultimately, I cannot reproduce what Ben has reported here.

Personally in the few captures I've made till now I definitely see a pattern, there's always a steep roll-off around 14-15 kHz plus an attenuation of 2-4 dB starting around 2-4 kHz. The latter seem to be kinda variable though, sometimes it's there, other times is slightly less.

And just to be clear, I always use the same exact setup and settings to make captures both with nam, tonex, tonocracy, aida-x, etc., and Tonex is the only one consistently having these issues... so it's not due to something imputable to my setup.

 

[Orvillain]

I think it is to do with the "mode" that you choose when starting your capture. I think the mode dials in a range of filters in order to reduce accuracy problems in their lstm implementation.

It could also be that lstm just isn't accurate when compared to the wavenet approach that NAM uses.

I also notice that ToneX tends to be a lot darker than the real amp, and I usually only ever do head-only captures.

When I did my MIR article on Kemper, QC, and ToneX, I did note that ToneX was much closer to the real amp than Kemper and QC. But it isn't as accurate as NAM.

 

[BenIfin]

Perhaps I could summarize where there is general consensus *

=> NAM is more accurate than Tonex across the full Freq Range
=> there is some sort of filtering going on in Tonex that is not going on in NAM or Helix Native or the Physical Helix Stomp Pedal
=> the degree of filtering - which varies - is evidenced in pretty much every kind of data plot we have seen so far
=> if the differences aren't audible to you and don't matter to you - its all good - probably use Tonex if you can - its very CPU light
=> if the differences are audible to you and do matter - its all good - probably use NAM if you can - its a huge CPU hog
=> IK's alteration to the marketing blurb here is very telling on their behalf
=> we all hear things differently so there is no one "right" format for everyone's ears

* - Yes I was wrong about the ~5k LPF cut, and amended the Thread Title and OP accordingly yesterday

 

[Orvillain]

* - Yes I was wrong about the ~5k LPF cut, and amended the Thread Title and OP accordingly yesterday

Your OP still says: It kicks in at around 5kHz

Which is why I brought it up.

 

[Orvillain]

PS: I think capturing is largely bullshit anyway. I'm so fucking over it.

 

[Ed DeGenaro]

Using the fundamental of the highest note on any guitar is not really relevant. The fundamental gives you no information about the harmonic structure of the pedals and amplifier distortion that come after the guitar.

I do not agree that inharmonicity (whether it actually is that, or not) is the same as aliasing. Aliasing necessarily involves wrap around frequencies, that ultimately ends up affecting and being perceived within the lower frequency range of the final tone.

Uhem…maybe re-read my post, you need the fundamental since it’s creates the partials. And the distortion adds partials according to it. Hence me saying the 13th and likely being completely unaudible would be at 16k.

 

[Orvillain]

Uhem…maybe re-read my post, you need the fundamental since it’s creates the partials. And the distortion adds partials according to it. Hence me saying the 13th and like completely not audible would be at 16k.

Gotcha.

 

[BenIfin]

Your OP still says: It kicks in at around 5kHz

Which is why I brought it up.

Thanks. My bad. I've now edited the Title and Thread and changed the content so it now just shows the 2 x RAW Aliasing Charts of Tonex and NAM at -90db with no cuts in either of them