Tonex aliasing

[banned]

I see

Yes, can I ask what the source of the aliasing would be? For instance, if they sample at 96K, you should be above Nyquist criterion....so hopefully you capture the source signal properly. Would the aliasing effects then somehow come from the modeling part? (which, they're applying a neural network approach and trying to get error below a threshold)

 

[DLC86]

I see

Yes, can I ask what the source of the aliasing would be? For instance, if they sample at 96K, you should be above Nyquist criterion....so hopefully you capture the source signal properly. Would the aliasing effects then somehow come from the modeling part? (which, they're applying a neural network approach and trying to get error below a threshold)

Aliasing is generated whenever there's an algorithm that creates distortion digitally, cuz harmonics created by the distortion are multiples of the signal and you need a much higher sample rate than what's needed for the audible range... that's why oversampling is used to mitigate this problem usually.
Tonex models run at 44.1 kHz natively (on the plugin at least).

PS: watch this video

 

[Richard_G]

So does this matter? I guess different people feel differently about it. But as far as I'm concerned, I don't appreciate that IK put "no aliasing whatsoever" in one of their six bullet points about the Tonex Tone Modeling Features when that's clearly not the case.

(And personally, I do think this matters beyond false marketing claims.)

 

[Richard_G]

With regards to aliasing, I don't have a clear understanding of the relationship between the software at the time of making a capture and the software later on using said capture. Is it possible to "cure" captures made early on of aliasing by later software developments? That is, would it be possible to reduce the aliasing with internal oversampling or is this not possible with capturing (or at least not with older captures)?

 

[blandified]

Aliasing is generated whenever there's an algorithm that creates distortion digitally, cuz harmonics created by the distortion are multiples of the signal and you need a much higher sample rate than what's needed for the audible range... that's why oversampling is used to mitigate this problem usually.
Tonex models run at 44.1 kHz natively (on the plugin at least).

PS: watch this video

Thanks for sharing. I was struggling to understand your plots, but the video explained it well.

If that's the case then, in your post that you linked, the FM9 looked the best right? Because the harmonics weren't being reflected back.

And also (I'm sure you've mentioned it, but I'm new to the topic and struggling to catch up) did you do the tonex/NAM captures yourself? I wonder if Tonex/NAM is like Kemper, where most people's captures aren't as good as the people (like M Britt) who have figured out the secret science behind it. Just thinking out loud though.

Both NAM/Tonex seem to perform well under Null tests. What is the relation between good null tests and bad aliasing?

 

[Mongillo19]

This is just weird petty shit, imo.

Spoken like a guy who has a Tone X on the way

 

[blandified]

It might not meet my standards. But.. if the ToneX doesn't (based on null tests) the other ones aren't likely to either. I think I will be happy with it.

And while I didn't like how the Helix sounded when I tried it last, before their major update, many people didn't care, so obviously some aliasing is not a deal breaker.

If someone is picking on the ToneX, they are picking on all the profilers (and probably many of the modelers.) If I had to guess, Fractal has the highest level of oversampling so it is going to be most immune to complaints about aliasing.

Its a slippery slope though. I wouldn't be surprised if some of the PC sims have even higher oversampling than the fractal.

I definitely hear you. Wondering how much aliasing translates to playing and hearing. I was quite happy with Kemper, but people came over and told me "stop having fun and look at these nulls tests." (obviously joking).

I would definitely like to see these aliasing tests done on Michael Britt profiles, I bet there would be significantly less aliasing and would look good in a null test since the whole capture process is a science/art that not many people have. We don't have the exact amps he used or the settings he used, so this isn't possible, but still would be nice.

I still want to know the relationship between null tests and aliasing artifacts. You'd think a good null test would mean a good aliasing test. But this is above my paygrade.

I think I'm team fractal now though.

 

[Orvillain]

I definitely hear you. Wondering how much aliasing translates to playing and hearing. I was quite happy with Kemper, but people came over and told me "stop having fun and look at these nulls tests." (obviously joking).

I would definitely like to see these aliasing tests done on Michael Britt profiles, I bet there would be significantly less aliasing and would look good in a null test since the whole capture process is a science/art that not many people have. We don't have the exact amps he used or the settings he used, so this isn't possible, but still would be nice.

I still want to know the relationship between null tests and aliasing artifacts. You'd think a good null test would mean a good aliasing test. But this is above my paygrade.

I think I'm team fractal now though.

Kemper had to address some amount of aliasing when it first launched, because people could hear very odd wolf tones and discrepancies in the high frequencies, particularly when playing high notes and double stop bends. It was very obvious back in 2011/2012.

I think it still does alias quite a bit tbh. But most people seem to be okay with where it is at right now.

 

[DLC86]

Thanks for sharing. I was struggling to understand your plots, but the video explained it well.

If that's the case then, in your post that you linked, the FM9 looked the best right? Because the harmonics weren't being reflected back.

And also (I'm sure you've mentioned it, but I'm new to the topic and struggling to catch up) did you do the tonex/NAM captures yourself? I wonder if Tonex/NAM is like Kemper, where most people's captures aren't as good as the people (like M Britt) who have figured out the secret science behind it. Just thinking out loud though.

Both NAM/Tonex seem to perform well under Null tests. What is the relation between good null tests and bad aliasing?

Yes, I've done the capture myself (captured an amp from the FM9) and also matched levels and gain to make sure to have a fair comparison.

People like MBritt surely know how to make good captures but that's due to taste and ability to dial in an amp and mic a cab. But no amount of those abilities can improve the aliasing performance (for a given distortion level), that can only be done by altering the code.

PS: aliasing can surely have an impact on null tests but it's not the only variable.

 

[DLC86]

@blandified forgot to answer the first part of your question... Yes, FM9 looks the best cuz it has proper oversampling (8x iirc), there's still a low amount of aliasing but it's inaudible, and you don't see the point of reflection on those lines cuz it would be way outside the graph (if it's 8x oversampled, they reflect back at 192 kHz)

 

[DLC86]

When you tested the ToneX, did you test the software or the pedal?

Software, I don't have a pedal... Anyway folks, pretty much all these questions and more are already answered in the original thread I linked above.

 

[jay mitchell]

Kemper apparently doesn't null as well as the others, but people still love it.

Null tests of nonlinear devices (guitar amps are always highly nonlinear) tell you very little. No modeler/profiler/capture device/plug can ever produce a perfect null for the range of signals a guitar puts out, so null tests are approximations. There's always a background level of error signal: noise, etc. Trying to set a single criterion - e.g., total error at -XdB wrt signal - does not account for the type of error. Random (white) noise is one type. Replication error (mismatch between linear and/or nonlinear device behavior) is another. Aliasing is yet another. The audibility - and therefore the relevance - of the types is not the same. Relatively large amounts of noise error in a null test may be irrelevant. Because noise is random, it will never null. Linear replication error - error in matching the linear frequency response of a device - would potentially be audible but is easiest to achieve a null with. Nonlinear replication error can potentially result in an excellent null at one choice of input level but much a worse null at lower or higher levels. Aliasing is audible at much lower levels than is random noise. Any amount of aliasing in a device will degrade a null test result, but a null test is a very poor means for quantifying it, as it can lie well below the level of random noise and still be audible and objectionable.

Claiming "no aliasing whatever" for the device used to make the clips in this thread is either woefully clueless or flagrantly dishonest.

 

[DLC86]

Also, having done some investigation on the Liquid Profiles, the ToneX seems to have a similar capability, in that it lets you choose the frequency range and Q of each EQ control. Which is largely what the liquid profiles are doing (you choose from a list the amp model you are profiling). With ToneX you have to actually figure out how to set those values.

Kind of... most tonestacks are passive EQs so controls interact with each other and it's quite hard to replicate their behaviour with a standard EQ... plus Tonex lacks the ability to place the EQ pre or post distortion

 

[DLC86]

Thats just not correct. Look at the ToneX manual, page 21 and you can see that any effect can be placed pre or post, including the EQ.

You're right, I always adjust it from the main controls on the amp and forgot there's a post button in the advanced controls... but that doesn't change the first part of my post.

 

[Foxmeister]

I know that aliasing was a big problem with Helix,

Personally I wouldn't say it was ever a big problem, but it is entirely subjective to the individual...

But before that, people were enamored with the Helix, so obviously it didn't affect majority of people's opinions.

See above...

It should be said that neither the Pod Go, nor any of the Pod Express pedals got the oversampling update that the Helix did, and you really don't hear of too many people complaining about squirrels with these products.

 

[paisleywookiee]

Someone needs to come out with a dummy-proof USB reamp box that makes capturing for ToneX and NAM as easy as the Kemper.

 

[blandified]

Pardon the noob question:

Will profiling/captures ever be able to "model" the amp?

Instead of capturing a single snapshot setting of the amp, can captures sweep through different amp controls and record mini-snapshots taken at each setting, thereby creating a "black box model" of the amp? I imagine this is impractical, as you'd have to sweep through every permutation of every single amp knob, perhaps making the capture-model too large to be used practically. The capture-model would have to know which knob is being turned too. Way too time consuming as well, unless there is a way of automating the process:

snapshot_000: volume1, gain1, tone1
snapshot_001: volume2, gain1, tone1
snapshot_002: volume3, gain1, tone1
snapshot_003: volume1, gain2, tone1
snapshot_004: volume1, gain3, tone1
snapshot_005: volume1, gain1, tone2
....
etc (going through all permutations)

Just thinking out loud. At this point just use the Fractal models (lol), but I like these thought experiments.

 

[Mongillo19]

Pardon the noob question:

Will profiling/captures ever be able to "model" the amp?

Instead of capturing a single snapshot setting of the amp, can captures sweep through different amp controls and record mini-snapshots taken at each setting, thereby creating a "black box model" of the amp? I imagine this is impractical, as you'd have to sweep through every permutation of every single amp knob, perhaps making the capture-model too large to be used practically. The capture-model would have to know which knob is being turned too. Way too time consuming as well, unless there is a way of automating the process:

snapshot_000: volume1, gain1, tone1
snapshot_001: volume2, gain1, tone1
snapshot_002: volume3, gain1, tone1
snapshot_003: volume1, gain2, tone1
snapshot_004: volume1, gain3, tone1
snapshot_005: volume1, gain1, tone2
....
etc (going through all permutations)

Just thinking out loud. At this point just use the Fractal models (lol), but I like these thought experiments.

Allegedly this is what the Quad Cortex models are. You can't make your own "complete" capture model but all the factory models on the QC may be exactly this.

 

[Mikael Dez]

Dummies aren't meant to capture.

Can confirm, am a dummy

 

[paisleywookiee]

Dummies aren't meant to capture.

That’s a dumb sentiment. This stuff should be easy to use, so it spreads out and makes the ecosystem better.