Tonex -vs- NAM ..... the Tonex "Aliasing" Low Pass Filter is at ~ 14kHz
- Thread starter BenIfin
- Start date
AFKAEjay(retired)
Roadie
- Messages
- 430
Not sure…should we phone the fire department?
BenIfin
Shredder
- Messages
- 1,086
What started me on this is that for the last 3 months I have been running an IPad Rig with Nam and Tonex and an IK iRig Pro Interface, Wah and Midi FSW's.
Was using Tonex iOS for Tonex Captures and Gigfast Lite for NAM Captures.
Hardware wise, the system is rock solid with seamless capture changes / spillover / efx switching all at 48k / 32 Sample Buffer - in short, whatever latency is there is "un-feel-able".
I use the *exact* same Amalgam Audio Tone King Tonex Captures and the *exact* same Amalgam Tone King Nam Captures - the are both captured at identical settings.
So I have one group of presets running via Tonex and one group running NAM to decide which way to go.
Once set up - with identical settings all round - 2 things were very clear after a day or so of A/B testing
=> the NAM Captures had audibly clearer, open and less constricted high-mids to highs
=> the Tonex Captures were audibly thicker and more congested in the mid-range
I thought nothing of it and just set about to EQ the Tonex Captures to match the NAM - no problems.
In short - no amount of tweaking / gaining etc.... got the Tonex patches to sound as good as the NAM Captures - and I spent a *lot* of time trying
I wont lie - my upfront very strong preference - was to want to stay with Tonex iOS as the CPU hit is 1/3rd that of NAM - hence why I spent so much time trying to get the Tonex Patchs to sound like the NAM Patchs.
Then I gave up. I just could not "fix" the difference and didn't know why.
Then quite recently DLC loaded his Aliasing Program and I started to wonder - based mainly on Cliff's comments about how poor the aliasing *can* be on AI/ML based Captures. took me a while with DLC's help to use it right but eventually got there.
In short, the tests above and in my other post visually confirmed the differences I could audibly hear.
They just don't sound and feel the same - so Tonex iOS is no longer on my IPad M1 rig.
Its totally solid - and am not hesitating to use it live.
F.w.i.w - this is the rig on my studio desk - very compact and light as well:-
Last edited:
James Freeman
Rock Star
- Messages
- 3,882
It looks like someone forgot to disable the LPF EQ in ToneX,
No conscious person that has anything to do with audio will put a permanent 5kHz LPF at the output to fix aliasing.
No conscious person that has anything to do with audio will put a permanent 5kHz LPF at the output to fix aliasing.
BenIfin
Shredder
- Messages
- 1,086
This is the software plugin - it can only be by design by the IK / Tonex Team - stuff like this just doesn't accidently get-through without being noticed or "deliberately allowed to be there" - all the above tests were done with Tonex 1.9 - the latest version which is now 2+ years since release.
The wont/cant change it now as every single Captures ever made would need to be re-done from scratch
Sadly, its exactly what you would do if you are trying to "mask" / "de-audible" as much as possible the aliasing in your product.
Last edited:
norminal
Rock Star
- Messages
- 6,652
Post in thread 'TONEX PEDAL ...... MAJOR FREE F/W UPDATE ..... Nov 2024'
https://thegearforum.com/threads/tonex-pedal-major-free-f-w-update-nov-2024.6590/post-267009
https://thegearforum.com/threads/tonex-pedal-major-free-f-w-update-nov-2024.6590/post-267009
BenIfin
Shredder
- Messages
- 1,086
Hmmmm ... I've thought about it - cant decide yet.
Would maybe cause a sh*tstorm with the Tonex / anti-NAM people and I dare say IK / Peter would respond with "pointless stuff" because I've been complaining over-there that they have effectively dropped Tonex iOS mainly because 2+ years later, you still cant load your own NAM Captures that you have bought into Tonex iOS ... unless you do a dodgy move and upload them to Tonenet and then into Tonex iOS and -the- remember to delete them from Tonenet.
Progressive_disclosure
Roadie
- Messages
- 129
That looks like a significant difference even to me!
The abandoned iOS app is one (of several) reasons I’m not interested in IK stuff anymore since it was a major selling point for me when Tonex was released. The last time I tried to use it on my phone it just kept crashing. Happily using NAM instead now.
Was the test done on the Tonex iPad app? Could the LPF be applied when using it? I always thought there was something off compared to the computer version, running the same captures.
The abandoned iOS app is one (of several) reasons I’m not interested in IK stuff anymore since it was a major selling point for me when Tonex was released. The last time I tried to use it on my phone it just kept crashing. Happily using NAM instead now.
Was the test done on the Tonex iPad app? Could the LPF be applied when using it? I always thought there was something off compared to the computer version, running the same captures.
BenIfin
Shredder
- Messages
- 1,086
Nope. All tests done on PC with Tonex 1.9 for PC DAW and NAM Plugin for DAW - Cakewalk was the DAW.
No Cab selected for either. Amp only
I tried several different LPF cuts and ~5k <-> ~6k was the one that replicated the Tonex Graphic in NAM the closest as per above.
No Cab selected for either. Amp only
I tried several different LPF cuts and ~5k <-> ~6k was the one that replicated the Tonex Graphic in NAM the closest as per above.
![[Nathan]](/data/avatars/m/0/613.jpg?1726885163){kind=avatar}
[Nathan]
Roadie
- Messages
- 567
I dunno what role this all plays with the capturing process but I have to believe it’s why tonex is so hit and miss for high gain compared to NAM. Kind of funny that 2dor and other capture legends figured out that doing captures with just enough high end will yield better results than setting things and not taking high end freqs into serious consideration.
northern_fox
Roadie
- Messages
- 472
People have said it is likely to be using a different model also (lstm vs wavenet) which could explain inconsistencies you mention moreso than aliasing
BenIfin
Shredder
- Messages
- 1,086
You're right - my ~5k estimate was probably "too harsh" - closer to ~14kHz in the following examples.
To ensure I did everything consistently, I just re-did the following 4 Graphs below just now. Everthing equally gained - no clipping - no Low/High Cuts, No filters etc...
a -> My RME UCX by itself as a test-calibration base
b -> Helix Native Latest PC DAW in Cakewalk - Placater Dirty - all Amp Controls on 10 - no Cab or IR
c -> NAM Latest PC DAW in Cakewalk - AmalgamAudio Carr Merc V S- Drive .txp - all Amp Controls on 10 - no Cab or IR
d -> Tonex Latest - PC DAW in Cakewalk - AmalgamAudio Carr Merc V S- Drive .nam - all Amp Controls on 10 - no Cab or IR
a -> My RME UCX by itself as a test-calibration base
b - Helix Native @ -40db Level No Low or High Passing
c - NAM @ -40db Level No Low or High Passing
d - Tonex @ -40db Level No Low or High Passing
I know *nothing* about how AI/ML Capturing works, but my take from the above 3 examples is that there is some "unusual filtering" going on with the Tonex "Capturing Method".
In these 3 examples, it looks like quite a steep Low Pass Filter at ~14kHz <-> ~15kHz in Tonex - whereas with NAM and Helix Native it appears (?) that no such Low Pass filtering is present.
F.w.i.w ... running all three of the actual aliasing tests at thresholds of -144db or -120db or -90db or -60db or -20db etc.... all reveal the same "trend"
I might be totally wrong - missing the point - am more than happy to be educated
Attachments
Last edited:
BenIfin
Shredder
- Messages
- 1,086
To quickly highlight the difference more "obviously" - and to "catch" as much aliasing as I could - the 2 tests below are at -120db - not only is the "filtering" or whatever is going on from ~14kHz and up quite obvious, but below ~14kHz, the "aliasing reflections " are significantly denser in Tonex
NAM at -120db
Tonex at -120db
NAM at -120db
Tonex at -120db
Last edited:
Mike Oliphant
Groupie
- Messages
- 60
I thought it was pretty well understood that Tonex does aggressive low (and high) pass filtering.
Not only does it mask aliasing, if you do it on the training signal as well (which I *think* I heard was the case, but I can't remember) then you reduce the complexity of the signal, likely increasing the accuracy of the resulting model for the remaining frequencies.
To be honest, it doesn't seems like an unreasonable tradeoff to make if you have limited CPU to run a model.
Not only does it mask aliasing, if you do it on the training signal as well (which I *think* I heard was the case, but I can't remember) then you reduce the complexity of the signal, likely increasing the accuracy of the resulting model for the remaining frequencies.
To be honest, it doesn't seems like an unreasonable tradeoff to make if you have limited CPU to run a model.
BenIfin
Shredder
- Messages
- 1,086
Probably one of the key reasons why NAM requires almost 3 x the CPU resources A/B compared to the exact same Captures in Tonex.
I was initially a very big and loud NAM skeptic, but I do genuinely hear the difference and have moved over to NAM from Tonex on my live rig.
When A/B'ing them the differences in the strength and clarity of the high end, and what I call the "congestion" in the mid-range, starts to become very obvious and clear as you get closer and closer to live-loud gig levels.
Ed DeGenaro
Shredder
- Messages
- 1,235
Why would you spend your time doing this actually? I mean the cab sim starts rolling of hard at 5k...
ian_dissonance
Roadie
- Messages
- 886
Because it’s more fun to find out how accurate your emulations are than to actually play guitar.
