Modelers and Aliasing

As an aside, if you're interested in ML researchers with SP backgrounds, check Joel tropp, Ben recht, Joan Bruna, Carlos Fernandez granda, Stephanie mallat.

I'll bow out now because I clearly don't know enough about the audio domain, eek.
 
Yes, you are right, they should, but they don't always, that was the point. You can test this for yourself with some of the JS plugins with insane levels of computer ending oversampling rates, given enough distortion, its hard to tell if the oversmapling is even doing anything. Lavry showed pretty convincingly that even 256 x oversampling would still leave in a lot of artifacts in the audible passband, though I argued that usually its not something you will actually hear, but its certainly there
 
I think that's just IMD plus beat frequency (the latter happens even on acoustic instruments cuz it's just interference between two nearing frequencies)
Why does the objectionable part of it seem to reduce with oversampling? Does IMD add a lot of higher harmonics which are more likely to result in audible aliasing?
 
Why does the objectionable part of it seem to reduce with oversampling? Does IMD add a lot of higher harmonics which are more likely to result in audible aliasing?
Since I know nothing about how and with which device/software you experienced this, I'll just leave this pic so you can draw your conclusions:


IMS-erf-img-1_637218740093807539.png



PS: but I guess that could be the case
 
How do you not add extra compute if you want to bake in an anti-aliasing filter... wouldn't it have to oversample to do so?
The weights in a neural network are just filters. The anti-aliasing filter is another filter. You can multiply the weights of those two filters and just store the resulting filter. No extra compute after the model is trained.
 
The input to an amp can easily reach ultrasonic frequencies. The initial pick attack has copious high frequency content.

Many distortion pedals have an output spectrum that reaches far beyond 20kHz.

If you limit the frequency to 10kHz it will sound dull.

None of that really matters though. What matters is the fundamental concept of aliasing which some here don't seem to grasp. Any time you have a nonlinear transfer function you create distortion. In guitar gear the transfer function is usually some sort of clipping behavior. At the limit this turns a sine wave into a square wave. A square wave has harmonics that extend well into the ultrasonic range.

If you don't oversample enough those harmonics alias into the audible spectrum. No amount of hand-waving changes that. The correct thing to do is to increase the sample rate and then downsample after all the processing is done.
The neural network could potentially learn to do anti aliasing without oversampling by virtue of the fact that internally it doesn't have only one signal but instead lots of features (potentially thousands). Whether or not you want to argue that that's effectively oversampling is another matter, but it's certainly already extremely redundant. I think the bigger issue is that ML experts for the most part are not signal processing or audio experts.
 
BTW @DLC86, the sweep should be 10k-20k with a quick fade-in/out at the ends to prevent a 'pop' in the measurement.

Use this sweep I made:
Code:
https://drive.google.com/file/d/1qM5OrM40K-RfQfm0K3mbPvl-VGMrPmqc
1Min 10kHz-20kHz Sine Sweep 48kHz, Mono and normalized to -0.1dB with Fade In/Out.
Lower its level in DAW.

My span settings, I click the span window to reset it before hitting play in DAW.
Offset Normalized so I can see the aliasing -dB value on the db scale on the right.
View attachment 5360


Big difference in aliasing with realistic vs unrealistic input dB values at the 10k-20k range.
View attachment 5361

Hi James !

Still trying to follow and hanging on by a thread :) ..... got your FLAC download and SPAN Plus and used your settings - thanks !

Got the FLAC ..... got SPAN Plus with your settings ..... am using Cakewalk by Bandlab

Am a bit confused about 2 quick things:-

a => should I use just an Amp Block by itself only in Tonex -or- should I also add a Cab / IR Block ?

b => cant see where / how do I adjust the Input DB Values ? - as for example you have in your test to say -10db and -35db ?

Many thanks again for all your help,
Ben
 
a => should I use just an Amp Block by itself only in Tonex -or- should I also add a Cab / IR Block ?

b => cant see where / how do I adjust the Input DB Values ? - as for example you have in your test to say -10db and -35db ?
A: Just amp block, no IR.
B: Any plugin with a volume control before ToneX will do.
 
Hi James / all !

Ok ... I must be doing something wrong ? :(

Playing - Listening Test

Playing at home, up loud [Gig Level] through my 2 x sets of NS10's [not at the same time] and then up loud [Gig Level] through my Celestion F12-X200 and Matrix GT800 Class AB Power Amp, and the Tonex, Helix and GT1000 all with the same IR .... the Tonex is noticeably / audibly clearer and more defined even than my GT 1000 - just sounds and feels better than both ... and the GT1000 is amazing as it is.

I recorded some samples of each and super critically listened repeatedly via both my monitors and my Audio Technica monitoring phones and there in *nothing at all* weird / strange / un-harmonic etc..... in any of the 3 platforms ..... even did multiple power chord hits and let them ring out to nothing ....... smooth as "hot butter" from intial attack to nothing ....... yet the 2 "graphs" below ... from what I can understand (?) ... seem to say there is a sh$t-ton of aliasing going on with the Tonex ?!?!?

Analytical Graph Test

I did two tests - (a) with the FLAC file provided and (b) with the Sig Gen provided ..... each set to -12db with the sig Gen at 10097hz .... results with SPAN Plus are:-

(a) Tonex Hi-GAIN Amp - No Cab - No IR -12db with 60sec FLAC File

1679136028511.jpeg


(b) [Same] Tonex Hi-GAIN Amp - No Cab - No IR with Sig Gen @ 10097 hz and -12db

1679136049753.jpeg


Any advice / suggestions / help would be good :) ... as "according to these graphs" the "aliasing" should be quite obvious !?!? and in short, the sounds are, for want of a better word, "perfect" in all ways ..... other than you may hate how I've dialed in the Tonex Amp tone :)

Ben
 
.... and just to make things "weirder" ...... below are the exact same tests with the exact same settings and setup ..... the only differnece is I am using the Placater Dirty in Helix Native .... jacked-up.

My real-world hearing and playing does not match with these Tonex vs Helix Alias results .... either I'm doing something wrong but if my testing is right:-

=> the Tonex is doing something else internally that these tests are not correctly measuring ?!?!?!?
=> the Tonex is maybe employing some sort of "filtering" or "masking" just before the output ?!?!?!?
=> something else altogether is going one !?!?!?!


Helix Native Placater - FLAC File Test

1679138652913.jpeg


Helix Native Placater - Sig Gen Test

1679138673976.jpeg


Ben
 
Last edited:
If you want to perform a meaningful test for aliasing - i.e., one that directly identifies audible aliasing artifacts - I described a method for doing so some years ago on a now-disappeared forum. I later described the test on TGP here: Digital Artifacts? Here's a quote of the relevant part:

"Using a DAW, create a sine sweep that goes from ca. 150 Hz to maybe 4k in 4-5 seconds and add harmonics. The application I use allows added harmonics up to fifth.

Apply that signal to the input of your modeler. If it creates aliasing products, from some point mid-sweep, in addition to the ascending pitch, you will hear descending tones. It's not at all a subtle thing. You can clearly hear aliasing that is 50-60dB in level below the fundamental."


Apply the test signal at a similar level to the one produced by your guitar. The test uses the most important instrument for detecting aliasing - your ears - and needs no interpretation. It enables you to clearly hear digital aliasing apart from other noise and distortion products.

Here's a link to the sweep I use for this test: Alias Test. If you're interested, it would be a good idea to download it ASAP, as I won't leave it up indefinitely.
 
If you want to perform a meaningful test for aliasing - i.e., one that directly identifies audible aliasing artifacts - I described a method for doing so some years ago on a now-disappeared forum. I later described the test on TGP here: Digital Artifacts? Here's a quote of the relevant part:

"Using a DAW, create a sine sweep that goes from ca. 150 Hz to maybe 4k in 4-5 seconds and add harmonics. The application I use allows added harmonics up to fifth.

Apply that signal to the input of your modeler. If it creates aliasing products, from some point mid-sweep, in addition to the ascending pitch, you will hear descending tones. It's not at all a subtle thing. You can clearly hear aliasing that is 50-60dB in level below the fundamental."


Apply the test signal at a similar level to the one produced by your guitar. The test uses the most important instrument for detecting aliasing - your ears - and needs no interpretation. It enables you to clearly hear digital aliasing apart from other noise and distortion products.

Here's a link to the sweep I use for this test: Alias Test. If you're interested, it would be a good idea to download it ASAP, as I won't leave it up indefinitely.
what took you so long seinfeld GIF
 
Thanks Jay. D/L the file.

Not using SPAN Plus or anything else - just listening several times - there was a very clear and very obvious audible difference between the two through my NS10's and Celestion F12-X200 and Matrix GT800 up at loud gig level.

Through the Tonex, the signal sounded quite "even and clear'ish" and "nice and smooth'ish and round'ish" with no noticeably "buzzing" for want of better words.

Through Helix Native it has a very noticeably lower pitch, and had a really annoying "buzzing" - especially in the lower to mid frequencies - according to my ears at least.

Nevertheless -way before doing any of these kinds of technical tests- the Tonex had a much clearer and more defined sound compared to my Helix Native and Helix LT ...... just sounded and felt better and "played" much more naturally - to me.

So I re-ran the SPAN Plus Graphs

Tonex Hi-GAIN at -12db

1679183022745.jpeg


Helix Native Hi-Gain -12 db

1679183081589.jpeg


Based on my very limited understanding and interpretations of these graph-plots, they seem to "verify" what I am hearing in the real world.

Thanks,
Ben
 
Not using SPAN Plus or anything else - just listening several times - there was a very clear and very obvious audible difference between the two through my NS10's and Celestion F12-X200 and Matrix GT800 up at loud gig level.
You missed the entire point, then. From my description above: "Apply that signal to the input of your modeler (or plug). If it creates aliasing products, from some point mid-sweep, in addition to the ascending pitch, you will hear descending tones. It's not at all a subtle thing."

I'll clarify: the sweep moves up in frequency. IOW, it begins at a lower pitch and moves to a higher pitch. If the system/plug/block you're evalating creates aliasing products of any consequence, you will hear pitches moving downward, IOW, going from a higher pitch to a lower pitch. Words like "clean," "clear," "smooth," etc., are completely irrelevant. They have no place in the discussion. Either you hear tones moving downward or you do not. In the latter case, there is no aliasing of consequence. In the former, there is aliasing.
 
You missed the entire point, then. From my description above: "Apply that signal to the input of your modeler (or plug). If it creates aliasing products, from some point mid-sweep, in addition to the ascending pitch, you will hear descending tones. It's not at all a subtle thing."

I'll clarify: the sweep moves up in frequency. IOW, it begins at a lower pitch and moves to a higher pitch. If the system/plug/block you're evalating creates aliasing products of any consequence, you will hear pitches moving downward, IOW, going from a higher pitch to a lower pitch. Words like "clean," "clear," "smooth," etc., are completely irrelevant. They have no place in the discussion. Either you hear tones moving downward or you do not. In the latter case, there is no aliasing of consequence. In the former, there is aliasing.

^^ That certainly wouldn't be the first time ;) :)

I understand about my wording being of no value now, as I clearly wasn't listening for the right thing -and- didn't understand / know what to listen for - I was only trying to describe the overall sound or tone I was hearing.

I will go back and re-listen closely and report back. Thanks again for the guidance.

Ben
 
UPDATE

Well .... that didn't take long at all ... and was a lot easier to do and understand ..... did it nice and loud through my usual NS10's and Celestion / Matrix combo and headphones ... all 3 up very loud ..... had both in the same track so was able to A/B them instantly against each other back and forth multiple times.

=> with both Helix Native -and- Tonex Native ..... to the best of my hearing .... there was absolutely no semblance of any kind of tones moving downward from a higher pitch to a lower pitch as the sweep wav file played all the way through ....even if it was somewhere deep in there, it was absolutely and completely in-audible to me.

So ...... hopefully my conclusion is right this time (?) when I say ..... neither Helix Native or Tonex produced any Aliasing of any "hear-ability" or consequence (?)

Really hope I've got this right this time (?) :)

Thanks again,
Ben
 
Back
Top