ALIASING is the worst enemy of our tone: How to hear it & how to fight it in NAM, Neural DSP plugin
Вставка
- Опубліковано 14 чер 2024
- In this video I would like to talk about the most strong and bad enemy of our beloved tone, the most relevant weakness of our amp simulator plugins, which is ALIASING. You won’t believe how much aliasing can ruin your tone. In this video we will talk about NAM, Neural DSP and Mercurial Audio plugins, verifying the different strategies we can use to deal with aliasing. Please subscribe and ring the bell to help me to make more videos like this!!!
▶ Follow this link if you want to purchase gear and help the channel:
- amp modelling pedalboards: thmn.to/thocf/xkwpquisc9
- pedals: thmn.to/thocf/hckhu3ukgo
- tube amps: thmn.to/thocf/dq0y6nqi99
▶💽 Link to a playlist of my songs: • Leo Gibson Songs
▶🎸🔉 If you are interested in my IRs, please follow this link: www.leogibsonguitar.com/shop-...
▶🎸🔉 If you are interested in my ToneX Profiles, please follow this link: www.leogibsonguitar.com/shop-...
▶ if you want to support the channel: www.paypal.com/donate/?hosted...
▶ Check out my first LEO GIBSON BAND song here: • Leo GIbson Band - #1: ...
▶ Check out my second LEO GIBSON BAND song here: • Leo Gibson Band- #2: O...
-----------------------------------
OTHER VIDEOS:
▶ Pedalboards demo review:
G2 x Four: • Zoom G2x Four / G2 Fou...
NUX MG-400: • NUX MG 400: demo and r...
NUX MG 30: • NUX MG 30: Demo and Re...
NUX MG 300: • NUX MG 300: Demo & Rev... BOSS GX 100 demo and review: • BOSS GX 100: Detailed ...
Valeton GP 200 LT: • Valeton GP 200 LT: dem...
Valeton GP 200: • Valeton GP 200: demo a...
Valeton GP 100: • Valeton GP 100: demo a...
Ampero Stomp II: • Ampero Stomp II: demo ...
Ampero Silver Edition: • Ampero Silver Edition:...
Fractal FM3: • Fractal Audio FM3: Dem...
Neural DSP Quad Cortex: • Neural DSP Quad Cortex...
HeadRush MX5: • Headrush MX5: Demo and...
Pedalboards Latency: • The most extensive amp...
Flamma FX-200: • Flamma FX 200: demo an...
Kemper Stage: • Kemper Profiler Stage ...
POD GO: • LINE 6 POD GO: Demo & ...
Zoom G11: • Zoom G11: Demo and Rev...
Zoom G6: • ZOOM G6: Demo & Review...
NUX MG 300: • NUX MG 300: Demo & Rev...
Mooer GE 300 lite: • Mooer GE 300 Lite: Dem...
Mooer GE 250: • Mooer GE 250: Demo & R...
Mooer GE 150: • Mooer GE 150: Demo / R...
▶ Modeling pedalboards comparison videos:
NAM vs Quad Cortex vs ToneX vs Kemper: • NAM vs ToneX vs Quad C...
Quad Cortex vs Fractal FM3: • Neural DSP Quad Cortex...
Quad Cortex vs Helix: • Neural DSP Quad Cortex...
Quad Cortex vs Kemper: • Neural DSP Quad Cortex...
BOSS GX 100 vs POD GO: • BOSS GX 100 vs Line 6 ...
BOSS GX 100 vs GT 1000 tech spec compared: • BOSS GX 100 vs GT 1000...
BOSS GX 100 vs GT 1000 sound comparison: • BOSS GX 100 vs BOSS GT...
HeadRush MX-5 vs NUX MG-30: • HeadRush MX 5 vs NUX M...
HeadRush MX-5 vs POD GO: • Headrush MX 5 vs Line ...
HeadRush MX-5 vs Ampero: • HeadRush MX 5 vs Hoton...
HeadRush MX-5 vs Mooer GE 250: • HeadRush MX5 vs Mooer ...
NUX MG 300 vs Valeton GP100 vs Mooer GE 150 vs Harley Benton DNA: • NUX MG 300 vs Valeton ...
NUX MG30 vs Mooer GE 200 vs Ampero One: • NUX MG 30 vs Ampero On...
FM3 vs Kemper vs Helix vs Pod GO vs NUX vs Valeton vs Mooer vs Soldano: • FM3 vs Kemper vs POD G...
BOSS GT 1000 core vs HX Stomp: • BOSS GT 1000 core vs L...
Zoom G11 vs Mooer GE 300: • Zoom G11 vs Mooer GE 3...
Fractal FM 3 vs Kemper Stage: • Fractal audio FM 3 vs ...
POD GO vs GE 250: • Line 6 POD GO vs MOOER...
GT 1000 vs Line 6 Helix: • BOSS GT 1000 vs LINE 6...
Line 6 Helix vs Headrush: • Line 6 HELIX vs Headru...
VALETON GP 200 vs POD GO: • Valeton GP 200 vs Line...
VALETON GP 200 LT vs HX Stomp: • Valeton GP 200 LT vs L...
VALETON GP 200 (LT) vs MOOER GE 300 (Lite): • Valeton GP 200 and GP ...
GT 1000 vs GT 1000 core: • BOSS GT 1000 vs GT 100...
GT 1000 vs FM 3: • Fractal FM3 vs BOSS GT...
GT 1000 vs HEADRUSH: • BOSS GT 1000 vs HEADRU...
GT 1000 vs GE 300 vs HELIX vs HEADRUSH vs PLEXI: • GT 1000 VS GE 300 VS H...
GE 200 vs GE 250 vs GE 300 vs GE 150: • MOOER GE 150 vs GE 200...
GE 250 vs Hotone Ampero: • Mooer GE250 vs Hotone ...
GE 250 vs HX Stomp: • Mooer GE 250 vs Line 6...
GE 250 vs Gigboard: • MOOER GE 250 vs HEADRU...
GE 200 vs Ampero One: • Mooer GE 200 vs Ampero...
GE 250 vs GE 300: • MOOER GE 250 vs GE 300...
GE 150 vs ZOOM G1X Four: x • Mooer GE 150 vs Zoom G...
BIAS AMP 2 VS AMPLITUBE 4 VS SOFTUBE VS FORTIN vs PLEXI: • BIAS AMP 2 VS AMPLITUB...
--------------------------------------------
⏰Timecodes:
0:00 - Intro
0:36 - Let’s hear aliasing
8:20 - Some theory about aliasing
10:45 - Thanks to…
#NAMf #NeuralDSP #Mercuriall
Aliasing happens, like you said, at frequencies above the sample rate. Analog to digital converters have anti aliasing filters which are very steep low pass filters. Any time you are processing audio, you run the risk of generating frequencies above what the sample rate is capable of, and therefore most of the time they use an anti aliasing filter.
Over sampling is converting the sample rate to a higher sample rate for the processing. This allows those higher frequencies to be processed accurately. But then it still needs to be converted back to the sample rate you are using, and a low pass filter is used. Over sampling gives you more accuracy in the processing, but the anti aliasing filter is the only thing that gets rid of the aliasing. That's why applying a low pass filter cleaned it up.
Also, keep in mind that any eq introduces phase shift. The more eq, the more phase shift. Test this by boosting a frequency at a narrow q on a very short click sound. If you boost enough, you can actually create ringing after the click. Linear phase eq minimizes this, but creates ringing slightly before the click and some after it as well.
This is also why its good to leave 0.7db of headroom when mastering your final mix of a song. When converting to mp3, the anti aliasing filter will cause some phase shift which can cause the volume to increase very slightly, and without that headroom you could clip your audio.
For more info, I recommend the book mastering audio by Bob Katz.
Thank you, really interesting info and points!
> Over sampling gives you more accuracy in the processing, but the anti aliasing filter is the only thing that gets rid of the aliasing. That's why applying a low pass filter cleaned it up.
The whole problem with aliasing is frequencies above the sample rate getting folded down into the audible range. You can't solve that by simply placing a digital low pass filter after the guitar plugin. Anti aliasing filters work in analog domain before the signal gets sampled. Once you're in the digital domain, there's no solution other than oversampling. Then there's the reconstruction filter, but that also happens in the analog and is handled automatically by the DAC.
@@boshi9 the digital low pass filter would happen before the sample rate conversion. Then nothing needs to get folded down, that's the whole point of the low pass filter, to eliminate those high frequencies.
@@boshi9 putting a low pass filter before or after the plugin may help, but only if it is getting rid of or minimizing those high frequencies that are causing problems.
By far the best channel for amp sim lovers! Thanks for the hard work.
You are welcome and thank you so much for this amazing compliment!!!
Leo is providing Jim Lill's quality content lately. Thanks for the video and I hope for loads to come.,
Thank you soooo much, really appreciated!!!
This is such a revelation Leo! Its like with computer graphics, when we have the same option "anti-aliasing" to prevent the "chopyness" of an image!! Great content as always! Keep on doing the awesome work!
Yes I do agree...it was a revelation also for me! It's exactly the same concept...thank you!!!
Thanks for putting this together. Much appreciate your work and efforts. 🙏
You are welcome!!!
Great video as allways. Thank you and greetings from Argentina!
You are welcome and I'm happy that you like the video!!!
Thanks again for providing some of the best and most useful information on how our gear works and the pros and cons. Your presentations, testing methodologies, and findings shed a considerable amount of light on topics that are often shrouded in technical complexities and audio superstitions. 👍😎
Thank you soooo much, really appreciated! I have to admit that this topic was obscured...It was fun to discover it and understand why sometimes we hear those strange artefacts...
Thank you!
You are such a good teacher Leo. I hadn't ever heard aliasing isolated like that before. Never really gave it that much thought before either. Great info as always. God bless and rock on 🎸😎👍
Thank you so much!!! Yes, I was also not giving too much credit...until I started listen carefully especially to the high freq...and diving in...well...it started to become clearer...
One would be misled by thinking your work is as correct as the headline executed. Thank you for disappointing.
I'm sorry I did not understand your point...could you please clarify?
@@LeoGibsonGtr Leo, the guy is being a troll. His comment was sarcastic. Press the "mute" button...🙄 Keep up the good work my friend!
😀👍
Leo your coverage of these issues like Aliasing, oversampling, latency, etc. are the best as you can explain them clearly, I like that you are focusing on these rarely covered topics... As for Aliasing, it can be pretty noticeable using distortion with sustain and pitch, like using a guitar with a Sustainiac and using the high overtones for sustain and you'll hear Aliasing a lot on cheaper amp sims
Hi, thank you!
Actually I did the video as I started hearing strange artefacts using a plugin...I never consider aliasing to be a problem, but when I started hearing that artefacts...well, I wanted to find the root cause ... and it comes out to be aliasing...it's such an important topic and so under evaluated....
@@LeoGibsonGtr Yes, unfortunately on iOS amp sims the aliasing is quite noticeable, not so much on PC plugins like the ones you used here
Leo, great video once again! This is a very interesting topic that I was eager to watch a good video on! You made it happen without me even having to ask hehe, thanks! This explanation not only brought some knowledge into guitar sim plugins, but also into the whole world of audio mixing and mastering as well (which I'm quite fond of)
Thank you sooooo much, I'm really happy to have been of some help!
Actually, while investigating this topic, I have learnt that aliasing is affecting some mixing plugins I was using...we never stop learning...
Wonderful video. You are so smart. I appreciate the time you take to research all this information.
Thank you sooooo much!!!
Another fantastic well explained video. Love it.
Thank you so much, I'm really happy that you appreciate the video!!!
Thank You Leo, always with the Best info! 🤘🎸
You are welcome!!!
Thanks for the knowledge! Great video.
You are welcome and I'm super happy that you like the video!!!
È proprio vero che "più si sa, più ci si accorge di non sapere"!!! Ottimo lavoro caro Leo! 👍
Grazie mille!!! Ho pensato la stessa cosa quando ho cominciato ad interessarmi al soggetto...lo avevo proprio sottovalutato...
Hi Leo! I love your channel and your videos! Greetings from Spain!
Hi Manuel, thank you so much for such a great compliment, I'm really super happy that you appreciate the channel!!!
love it! thanks for the knowledge :)
You are welcome and I'm happy that you have appreciated the video!!!
Great video! I appreciate the info :)
Thank you so much, I'm really happy that you like the video!!!
Something else to consider is intermodulation distortion (IMD). It’s more of a problem with sample rates above 48kHz and broadband signals like mixes as opposed to guitars or basses which have limited high frequencies. Still, filtering helps there also and that’s what the Ultrasonic plugin from TDR deals with. I find a regular LPF ahead of each nonlinear stage can clean up a sound.
Love your videos. I did this test on the Ampeg SVT Suite plugin and it was great. I would say it was better than the Nural DSP or NAM. But when I did it with the Bogren Bass Knob, WOOOOOOW!!!!!! And not in a good way even when it's set to run as clean as it can be. I also tried it with the Waves SSL EV2 plugin and it was FANTASTIC.
Thank you sooo much, I'm really happy that you like the video!!!
Thank you also for sharing your experience!!!
@@LeoGibsonGtr I was playing with the low pass trick and, although the Ampeg SVT Suite was already good, it did improve it. I haven't tried it with my actual bass yet, so I don't know what it's doing to the tone. Seems about 10k is the sweet spot. The Bass Knob is better, but still pretty bad even down to 7.5k.
It's not surprising that an EQ / channel-strip plugin exhibits much less aliasing than a distortion / amp plugin. This is to be expected due to the nature of the plugin processing.
Well. Who knew? Great video. Thank you.
You are welcome!!
😎 thanks 😊
Thank you!
One of more interesting video that I never saw...
Thank you, I'm really happy that you have found the video interesting!!!
wow, nice Picture Quality!
Thank you!
Another great video, thank you Leo. You might want to change the title of the video from "worser" to "worst". I know English is not your first language, just trying to help. Also there's a new version of NAM out now.
Thank you so much, both for the compliment and the suggestions!!!
The problem is that aliasing can easily be addressed by using oversampling / anti-aliasing, but in guitar amp plugins you want good realtime performance without latency.
Since oversampling / anti-aliasing creates latency and also higher CPU demand, most developers don't put it into their guitar amps.
A clever solution would be a switch that oversamples the amp plugin whenever you "bounce" (export) the audio (for example during the final mix and mastering stage). However, some DAWs like Ableton Live do not fully support this method called "offline oversampling".
Hi, thank you for sharing!
Let me add:
1) there are some plugin that allows for oversampling only while exporting your project, like the DiBi quadro
2) some plugin does not work with oversampling, for instance NAM
Thank you Leo for a very clear demonstration of the degradative effects of aliasing on guitar tones. What sample rate are you recording in, and would it be possible and beneficial to work at a higher (96kHz) sample rate. I know ToneX is locked to 44.1 or 48 kHz or something. Does NAM allow you to create profiles working at the higher rate?
You are welcome! I always work at 48 kilo hertz. It would be beneficial to work at higher frequency and this is what actually overdamping does, even if you are gonna tax a lot the CPU and you are gonna need much more space in your hard disk. Nevertheless with NAM you can not create profiles higher than 48 kilo hertz...in order to do it you should change the source code.
Thank you!
@@LeoGibsonGtr Thank you Leo.
👍
Hi Leo and thanks for your content, in this post case about plugins. I use NAM just as a beginner. Not the profile side of it but just the plugin. I see it as a new amp, or many new amps ;) just downloaded a few well crafted profiles and use it standalone. As an amp, simple like that. And I have managed to make it work on legacy windows 7 system. On win 10 it just shuts itself down and have read in the bug tracker that someone is experiencing the same on win 11. My question is: are you experiencing the same? Cannot find a workaround to make it usable on win 10 so far...
Hi, you are welcome! I have used with an Apple Mac and in windows 12 without experiencing problems. I have not tried in win 10, therefore I cannot tell if it works or not.
Thank you!
Hi Leo, are you aware of the aliasing function in Aida X? It’s set to 66% as default. Not understood it until just now so not had a chance to assess this function in Aida X
Buonasera GRANDE LEO...Ho il Tonex Pedal... Ho notato che per i profili di IK MUltimedia sulle alte frequenze è tutto regolare...per i profili di alcuni utenti invece noto aliasing..... da cosa dipende? Spero che la mia domanda sia consona.... Grazie
Ciao,
è un'ottima domanda. Non sono però sicuro della risposta...Personalmente non ho fatto analisi specifiche sui profili fatte da me, ache se non mi sembra di notare aliasing invadente...Ciò detto, se si usa il ToneX come interfaccia audio, essa lavora a 44.1 khz, mentre sulle specifiche tecniche viene data a 192khz...presumibilmente quindi lavora in oversampling....dopodichè entrano in gioco i convertitori usati per fare i profili, che potrebbero non gestire bene l'aliasing...anche se credo che tutte le interfaccia audio moderne abbiamo filtri antialiasing.
Topic interessante comunque...se riesco nelle prox settimane proverò a fare delle verifiche.
Grazie
Why oversampling the NAM with DDMF doesn't work (4:49)... I think it's because the sample rate NAM's IR conflicts with the oversampled rate. Perhaps using a higher sample rate on your interface/DAW might give different results, or choose the IR with a higher sample rate when oversampling.
Amazing, now, does this problem also exist on the quad cortex amp captures and preset amps?
Thank you!!! I suppose it does...but I'm not quite sure...I need to find a way to test it..
Thank you!
Thanks dude! If you still have the fractal stuff would be interesting to have an overall comparison with QC :)
You are welcome! Yes, that is another thing to test...even if I would have to go fully digital to take the measures, otherwise the anti aliasing filter of my digital audio interface would reduce aliasing...thanks!
In the NAM example, the low pass filter after the plugin cleans up the sound nicely. Would the same technique apply to other amp sims, or would you want to use a low pass filter before the amp sim to help prevent aliasing?
Hi, the low pass filter (in my opinion) has to be applied after the plugin, as the aliasing is produced by the plugin (due to the fact that for distorting the tone, the plugin produces overtones at a freq which is higher than the Nyquist one). Yes, it applies also to other plugin if they produce a noticeable hiss or noise in the high freq.
Thank you!
NAM works properly only at 48kHz, maybe in future releases will support with new models trained to another frequencies. So when you run NAM in metaplugin with oversampling, it still works in 48kHz, ignoring frequency doubling.
Yes, exactly, we need that also the training procedure is step up at 96 kHz.
Aliasing cannot occur just a little bit, if it is there everything is completly distorted. Aliasing only occures when you sample freqs higher the Nyguest theorem.
Will plugins suffer from aliasing if you use analog gain pedals into a low-gain plug in? Or does the aliasing only occur when the plugins themselves produce too much gain
As far as I know, the more gain you obtain "digitally", the more aliasing you can introduce. Therefore the solution you are mention should reduce aliasing, also considering that typically audio interfaces have anti aliasing filters.
Thanks!
I'm quiet happy with DDMF's Metaplugin and Grand EQ. I think some plugins (e.g. NAM) don't allow Metaplugin to oversample them due to non-standard coding. I've seen Metaplugin even crash out in some plugins - not just glitch out the audio. Usually it works great though, and it can surprise one that even plugins without known saturation can be improved, like synth plugins. (Some synths have distortion effects in them but even those that don't claim to include these still might be helped - set it to 8x and visualize it on an EQ and you may see some change). Harmonic content is somehow becoming aliased perhaps? Its still a bit of a mystery, but if it sounds (and looks on graph) like a big improvement I just go with it.
Hi, thank you for sharing!
Yes, I think that if a synthesiser produces overtones, then aliasing may occur.
How much frequency content does a guitar signal have in the over 24k range?
Hi, the problem is not related to the guitar signal, but to the distortion or saturation you apply. When you apply distortion or saturation to a signal of (for instance) 8khz...the distortion generates many overtones that go beyond the sampling rate you are using.
Thank you for your question.
Aliasing can be explained as the reflection of sound from the upper frequency limit, everything above it is reflected back.
Yes...and distortion produces harmonics above that threshold, that therefore are sent back as disturbing noise. That's why aliasing is important for us guitarist..as we use a lot of distortion...
Thank you!
I was never happy with the results of different modelers, to get a sound that I liked I always used in low pass filter between 6000 and 9000, now I understand why.
Hi Diego, thank you for sharing!
AIDA-X has an anti-aliasing adjustment directly in the plugin 🙂
That's correct....thanks
A neural network can only work at the frequency at which it was trained, aliasing can only be avoided if the developer supports 96 kHz and it can be cut off above 24 kHz with a filter, I really look forward to this.
Hi, thank you for sharing. Here there are maybe some other ways to avoid aliasing: github.com/sdatkinson/neural-amp-modeler/issues/221
Is aliasing only exists in 44.1k sample rate project? I checked NDSP plugins through PluginDoctor with 48k sample rate and I saw the clear screen, but I didn't check it with sine.
Hi, aliasing exists also at 48 and even at 96...By the way, in this video all the tests are done at 48 khz.
Thank you|
@@LeoGibsonGtr got it, thanks!
You are welcome!
@@LeoGibsonGtrbut is it less?
you can hear aliasing by simply bending a note. because of this comparing a black box model like NAM in A/B would show these artifacts. aliasing with black box modeling is an issue. not sure how this will be solved. in white box modelling this can be fixed by oversampling indeed
Hi Marco, you mean that white box can oversample and black box methodologies cannot?
Thanks!
@@LeoGibsonGtr If I am correct (I am no DSP guru) oversampling needs to be done when training the model. So this is different from the white box models where you can simply apply oversampling as part of the DSP. NAM works at the moment at 48 kHz if I am correct. So that means: no oversampling at all.
Yes, I think you are right...
Thank you!
@@LeoGibsonGtr you're welcome
👍
This has been around since the first CDs. Lowpass filters remove the high frequency "distortion" but doesn't help the "missampling" that occurs by not having a high enough sampling rate to accurately reproduce the waveform.
The oldest explanation was a 44khz sampling rate, the standard for CDs, can sample a 20khz twice. That's enough to give the frequency but not the shape of the wave. It can't tell if it's a sine wave, square wave, or any other shape of the wave. Just that it's 20khz. And as we know a square wave sounds nothing like a sine wave. I thought that the sampling rate of modern digital audio sources had been increased to eliminate this though? It's been over 40 years since this was discovered. Why is this still an issue?
HI, it is reduced of course by our audio interfaces when converting, but it is a consequence of the Nyquist theorem...I think it will be solved when we will use for instance 256 kHz of sampling frequency ...or something like that...it's just an opinion obviously. In order to use such an high freq we need very powerful computers and a lot of hard disk space...otherwise we would have to increase a lot the buffer size, with a lot of latency....
@@LeoGibsonGtr My point was that this is a known limitation of the 44 khz sampling rate and they have had oversampling to help this for decades. Sure it requires better hardware, but it's out there.
Use LSTM instead of WaveNet with NAM. Done.
Over-sampling after the signal is physically sampled is nothing more than interpolation. The real over-sampling is when the ADC is actually set to over-sample at a higher sampling rate than actually required. In your application, the best practice is to use an anti-aliasing filter which is simply a low-pass filter before digitization. However, if your media is already sampled and digitized (a wave file for example), it is practically impossible to apply this technique either. For guitarists using digital effects processors and amp simulators: simply role back the tone knob a bit before recording anything, that will act as an anti-aliasing filter, or just place an EQ early in the signal chain rejecting anything above 8Khz.
Hi Alex, that's basically what I'm suggesting (if I properly understood your suggestion) in the video. I have used a low pass filter at 7.2 khz (in this specific case that was the sweet spot for me) in the chain just after the NAM plugin.
Thank you!
@@LeoGibsonGtr your channel and factual explanations are the best thing I've seen on the internet about audio in a scientific way!!!! Many thanks Leo!!!! One question, in this example you are placing a lowpass filter AFTER NAM, but as far as I understand @alexm66 is suggesting to low pass the guitar DI track BEFORE going into NAM???
A-liasing, not Ah-liasing. Like in Hey, not Hi
Thanks
@@LeoGibsonGtr no worries! I'll have to say this video has been very useful, as I've used the tests to test out a program and have found that even though it states it oversamples that it actually does nothing.
wouldn’t a good high order low pass filter at 20k before the plugin fix this , for anyone over 25 you can even go down to 18k and you probably won’t hear any diference 😂😂. With NAM it seems to double the overtone frequincy when you add it to metaplugin, lol.
I always comment before I watch the whole video, try using the lowpass before and after in the super high part of the spectrum around 20k . Nam works at 48k so has a theoretical limit at 24khz, so it should be very good in the 20-20k if you filter in and out. As a side note it might also be good to use filters throughout the sound capture for the models , so the sistem never sees anything above 20k in it’s entire existence in case some of the high frequency artefacts might come from the training data aliasing.
Hi, yes, the low pass filter is what I'm also suggesting in the video: if you watch the video you can see what is the freq I suggest to use. Nevertheless this does not entirely solve the problems, as the folding back freq as also below the low pass filter.
Thank you for your message!
@@LeoGibsonGtryes but you used the low pass filter on the output,( i think, maybe I was not paying attention😅), I was saying you can use it on the input to prevent the high frequency above 20khz causing aliasing in the model, and keep more of the higher harmonics in the output. anb because it’s a guitar you can probably go lower than 20k.
It might also answer the question if it’s the model or how neural networks work, if you remove signal above 20k there is no possible aliasing since nothing ever touches the higher end frequencies. If you still have those artifacts then it’s the model.
Hi, I don't think it's gonna work, as aliasing is not produced only by high frequencies. The distortion effect creates overtones also for lower freq...that then goes beyond the Nyquest freq and fold back.