thegame
Well-known member
But will A2 be a happy meal in 3 years?
CrystalSky
Well-known member
Next iterations will probably come and remain at the top. Open source is the king.
Shask
Well-known member
Drew tends to think any sound is good enough as long as you can record or perform live without issue. Accuracy is not important compared to something that makes enough noise to sound like a guitar.
Sound Accuracy and how you use the sounds are two different conversations, IMO.
Shask
Well-known member
One nice thing about NAM is if you keep the audio files, you should be able to retrain your sounds to the new version when it becomes A3.
C
chumbucket
Well-known member
I didn't really notice a difference between A1 and A2 from an audio quality standpoint, the most noticeable thing was that the CPU load went down significantly.
thegame
Well-known member
Can you expand on that please?
Shask
Well-known member
Between ToneX V2 and NAM A2, they are both very good, but if you just let a chord ring out, I notice NAM has more note clarity. It is a little more hi-def, less grainy, more clear, etc...
CrystalSky
Well-known member
NAM A2 allows you to make a capture not of a single knob position, but somewhat the whole spectre of them. @2dor can give you more detailed info because I am broke and don’t have first hand experience.
Last edited:
C
chumbucket
Well-known member
Along those lines, I think that NAM beats Tonex when it comes to backing off on your volume control too. It behaves more like the real amp does.
Shask
Well-known member
I dont do that much, but I can believe it. It is those fine details that takes the new capturing tech from great to amazing. The pick attack is the area I am picky about. Most classic modeling devices have a squishier smooth pick attack that bugs me. I dont notice that with captures.
N
Nigel
Well-known member
I hung out with a friend who's a great in the box guy and we loaded up the NAM plugin and a basic bitch Vox AC30 Top Boost model. It sounded (and FELT) like I was hearing a Brendan O'Brien level bed track.
2dor
Well-known member
That was also possible with NAM A1 but a lot less people had access to a framework that would enable it.
Coincidentally, Two Notes released Genome 2.0 around the time when NAM A2 launched and they partnered with Artera DSP to includ a framework / app. (called Capture Studio) which walks users through capturing & training parametric NAM models of their gear; it uses NAM A2 under the hood, yes.
In my tests, it performs best when you don’t have too many knobs to model and has some limitations (ie. one Gain knob; so Plexi models with Vol 1, Vol2 might turn out very weird etc).
There's recently been an open-source framework published on Github which does this a little better (and it's free); it covers a lot more of the amp's control space (knob permutations), allows you to use your own training and validation signals and has some very nice QoL features like knob range (useful for Rectos where anything above 3 or 4 is mushy), step size, avoid 0 etc.
It's here if anybody wants to check it out:
https://github.com/phillipmself/neural-amp-modeler-parametric
Still needs some polish but it's a huge stride in the right direction.
The models it produces can be loaded in this free plugin which will read and display all knobs from the trained model:
https://github.com/phillipmself/NamParametricPlugin
This plugin doesn't have support for calibration metadata yet but the author's logged it as a future feature.
Last edited:
2dor
Well-known member
The "calibration" bit tackles a different problem which is as follows:
A challenge seen in early-days NAM (and amp sims suffer/suffered from the exact problem) was people were getting pretty different results when loading the same NAM capture / profile / model.
Your hard-rock JCM800 profile would sound clean for one user while another user would report that it's straight up unusable because of all the crazy amount of gain it had.
This came down to audio interfaces and how their instrument inputs are not equally "hot" across different manufacturers (and that's not even accounting for users jacking up the Gain knob on their Guitar Input haha).
If you were to generate a 1KHz 1Vrms sine tone into the instrument inputs on interfaces from different manufacturers (ie. Audient, Focusrite, RME, MOTU etc), chances are the resulting dBFS readings you'd see in the digital domain / DAW would not be the same.
And it's this exact thing that calibration wants to tackle.
It splits the responsability between:
1. the NAM profile creator: needs to ensure they know what their signal chain is calibrated for by measuring the maximum voltage value (with a multimeter) their interface puts out as the signal reaches the amp's input during the reamping process. That's the short of it; the long read is here:
https://neural-amp-modeler.readthedocs.io/en/latest/tutorials/calibration.html
2. the user or hardware / pedal manufacturer: need to ensure they know how hot their guitar instrument input is and disclose this value to the software player they're using.
For people using the vanilla NAM Gateway plugin (which is still probably the best choice for anybody starting with NAM), the guide I did last year still holds up as far as getting started (as an end-user) with calibrated profiles:
Having a good grasp on & using calibrated NAM profiles allows users to get an accurate response from stacking several NAM profiles together like:
- NAM boost > NAM amp
- NAM boost > NAM preamp > NAM poweramp
The vid that walks people through stacking calibrated profiles is below; recorded for A1 but still applies for A2 as well:
Sorry for the bedroom video production etc; I'm not a YT guy - just did it to help folks out.
- How do I, the author, make sure that this profile behaves exactly as the real amp I'm modeling across a multitude of audio interfaces / units ?
A challenge seen in early-days NAM (and amp sims suffer/suffered from the exact problem) was people were getting pretty different results when loading the same NAM capture / profile / model.
Your hard-rock JCM800 profile would sound clean for one user while another user would report that it's straight up unusable because of all the crazy amount of gain it had.
This came down to audio interfaces and how their instrument inputs are not equally "hot" across different manufacturers (and that's not even accounting for users jacking up the Gain knob on their Guitar Input haha).
If you were to generate a 1KHz 1Vrms sine tone into the instrument inputs on interfaces from different manufacturers (ie. Audient, Focusrite, RME, MOTU etc), chances are the resulting dBFS readings you'd see in the digital domain / DAW would not be the same.
And it's this exact thing that calibration wants to tackle.
It splits the responsability between:
1. the NAM profile creator: needs to ensure they know what their signal chain is calibrated for by measuring the maximum voltage value (with a multimeter) their interface puts out as the signal reaches the amp's input during the reamping process. That's the short of it; the long read is here:
https://neural-amp-modeler.readthedocs.io/en/latest/tutorials/calibration.html
2. the user or hardware / pedal manufacturer: need to ensure they know how hot their guitar instrument input is and disclose this value to the software player they're using.
For people using the vanilla NAM Gateway plugin (which is still probably the best choice for anybody starting with NAM), the guide I did last year still holds up as far as getting started (as an end-user) with calibrated profiles:
Having a good grasp on & using calibrated NAM profiles allows users to get an accurate response from stacking several NAM profiles together like:
- NAM boost > NAM amp
- NAM boost > NAM preamp > NAM poweramp
The vid that walks people through stacking calibrated profiles is below; recorded for A1 but still applies for A2 as well:
Sorry for the bedroom video production etc; I'm not a YT guy - just did it to help folks out.
Last edited:
N
nightlight
Well-known member
Paradex is NAM A2, but is a method of capturing an amp with the knobs at all possible pictures. Still a work in progress though, it is in Beta testing right now.
C
chumbucket
Well-known member
^This.
Input calibration is so key to getting good results with NAM. I imagine that there are a bunch of folks out there who've tried NAM without adjusting their settings, and have brushed it off as not sounding all that great and completely dismissed it.
N
Nigel
Well-known member
Fantastic!
2dor
Well-known member
It sure is - a lot of folks had that happen with amp sims too until @easstudios did his research / data gathering on calibration levels for them.
I know, for a fact, quite a few folks have tried NAM without paying attention to calibration, brushed NAM off and then couldn't stop singing its praises after they used it properly.
rsm
Well-known member
I've been using various digital modelers for many years, starting with the first Vox ToneLab, the tabletop version, currently using a Helix Stadium XL as my modeler, and a VP4.
input gain / buffers were an issue back when I had my G-System too; and just about every digital modeler I've used.
thinking back on guitar technology of the last 20 - 30 years, it's amazing to consider the collective time, effort, investment, etc. just to recreate the sound and feel of tube amps and analog signal processing; mid-20th century tech.
I suppose once you understand how captures work, and have your capture "rig" configured, making good captures becomes easier? just as learning how your digital modeler works and how to get the sounds you want takes time to learn and become proficient makes it faster/easier.
It's still not as easy as turning on a good tube amp you like, with whatever effects pedals/racks you like.
Digital ease of use has come a long way, but it's still not tube amp simple. Tube amps just do what they do; what they've always done; simple and easy.
SMH.
input gain / buffers were an issue back when I had my G-System too; and just about every digital modeler I've used.
thinking back on guitar technology of the last 20 - 30 years, it's amazing to consider the collective time, effort, investment, etc. just to recreate the sound and feel of tube amps and analog signal processing; mid-20th century tech.
I suppose once you understand how captures work, and have your capture "rig" configured, making good captures becomes easier? just as learning how your digital modeler works and how to get the sounds you want takes time to learn and become proficient makes it faster/easier.
It's still not as easy as turning on a good tube amp you like, with whatever effects pedals/racks you like.
Digital ease of use has come a long way, but it's still not tube amp simple. Tube amps just do what they do; what they've always done; simple and easy.
SMH.
Shask
Well-known member
I need to figure how to calculate the number to type into the capture for calibration. I haven't done that, but I have my reamping setup dialed in, so I could easily add it.
