• TalkBass has been independent since 1998. Add your voice.
    Create a free account to reply to discussions, view embedded media, and browse with fewer display ads.
    Join free Log in
    Want zero display ads or expanded classifieds tools? Compare plans.

NAM (Neural Amp Modeler) big thread

#101
gtirard said:
I've just finished capturing my amps, preamps and pedals:

www.tone3000.com

gtirard's NAM profiles & IR's · TONE3000

Explore gtirard's latest Neural Amp Modeler (NAM) profiles and impulse responses (IR's) for guitar tones, bass tones, recording and more! · I'm the bassist of InRed - an indie rock band based in Paris (France). I play with a pick 70% of the time. I don't use many pedals, but I like tube amps and...
www.tone3000.com

Feel free to comment (some of them may need improvements)
Click to expand...
I guess I should have a go.

The ones I have of interest are probably the JD-10 pedal (thanks, Neil Murray) and the Boss OD-2 is relatively uncommon and I can run that with both boost options and more than one power supply. I'll probably run through the Award-Session recommended settings from the manual.

I should have capture my Warwicks before I sold them.

I like the Behringer V-amp synth sounds, but they are not capturable. New word...
 
#102
OD-2 - tone low - drive low - turbo off
OD-2 - tone low - drive 9 o'clock - turbo off
OD-2 - tone low - drive noon - turbo off
OD-2 - tone low - drive 3 o'clock - turbo off
OD-2 - tone low - drive full - turbo off
OD-2 - tone mid- drive low - turbo off
etc

It takes a while just to capture some of the options... Let alone train it all.
 
#103
gtirard said:
I've just finished capturing my amps, preamps and pedals:

www.tone3000.com

gtirard's NAM profiles & IR's · TONE3000

Explore gtirard's latest Neural Amp Modeler (NAM) profiles and impulse responses (IR's) for guitar tones, bass tones, recording and more! · I'm the bassist of InRed - an indie rock band based in Paris (France). I play with a pick 70% of the time. I don't use many pedals, but I like tube amps and...
www.tone3000.com

Feel free to comment (some of them may need improvements)
Click to expand...
Thanks so much!

Will grab it all today and check it out.

Loads of cool stuff! Thanks for adding to the community.
 
  • Like
Reactions: gtirard
#104
-Asdfgh- said:
OD-2 - tone low - drive low - turbo off
OD-2 - tone low - drive 9 o'clock - turbo off
OD-2 - tone low - drive noon - turbo off
OD-2 - tone low - drive 3 o'clock - turbo off
OD-2 - tone low - drive full - turbo off
OD-2 - tone mid- drive low - turbo off
etc

It takes a while just to capture some of the options... Let alone train it all.
Click to expand...
I'm about 1/3 way though captures of two pedals... I'll try to remember to noise gate before rendering and uploading...

I thought that the OD-2 was dead until I realised i was using a 9V not 12V supply - the pedal works in a starved mode at 9V but is suppose to be run at 12 and then the LED lights enough to actually be visible. It's not pedal that is of any value for bass, though.
 
#105
gtirard said:
I do think you're the perfect candidate indeed, at least if you're always using the same EQ settings on the Peavey head - make one capture for each favorite EQ setting and call it a day. If you are used to tweaking the Peavey's weird EQ each time you use it, you won't find a virtual reproduction of the Peavey's EQ on the Dimehead or the Anagram.
Click to expand...
yeah, nope, I've got them split between highs and lows with a crossover but would be fine with the aforementioned weird EQ being applied to my entire signal chain and I only adjust them slightly based on the room assuming I can't get a good result from adjusting it on my basses, which I usually can.
Thanks! There's a weekend project for me.
 
  • Like
Reactions: gtirard
#106
Onto the JD-10 captures. I may try the Guv'nor 2 (with snipped capacitor mod). I gave away my original to a friend who used it. The JD-10 and original Guv'nor are viable for bass. I don't much like the Guv'nor 2 to be honest. I nearly got a Daddy-O back in 'the day' but won't at current prices.
 
  • Like
Reactions: DaBigBadaBoom
#107
I ran out of free time with a GPU on Collab and my GPU is AMD so not supported. I will have another go and trying to get PyTorch to run on AMD and integrate it in to the NAM command line, but for the moment I'll just let my PC grind away for a couple of days using just its CPU. A GPU does make a huge difference on training times. I could fire up an Amazon EC2 instance, but that will cost me money, although so does running my PC. My PC has lots of cores available but I don't know how many it is using as my desktop PC is running Windows for the available music software but really I am a Linux guy so I don't know how I'd check on parallelisation without going down another rabbit hole.
 
#108
I recorded the OD-2 and JD-10 but I am not sure what I did wrong as it refused to train them. So I deleted them. But it had used up my free GPU time, it seems, doing that. So I am doing the Marshall GV-2, which I hadn't tried with bass, but does seem to work. If anything, it works better for bass than guitar, even with the C21 capacitor removal mod. Anyway, those are training on the CPU and I will come back to the others. I'll upload to tone3000 when done. Don't wait up.
 
#110
gtirard said:
I've just finished capturing my amps, preamps and pedals:
Click to expand...

I downloaded and tried a bunch of these, thank you for the effort in capturing and uploading! I assumed the Monique capture was going to be my favorite because I have a Simone, but mon dieu that Jad Freer Capo! Is that really what they sound like? I’m floored. I’ve always wanted to try one but they’re pretty scarce out here in the middle of the ocean. I think I need to build a patch around the Capo.
 
  • Love
Reactions: gtirard
#115
I played around to see if alternative 'input.wav' files are possible and the answer is no. I may need to delve into the internals and fork to (a) add AMD support effectively (b) see if I can use different input files to see if I can train other relationships other than what is effectively probably a series of input-dependent Fourier transforms, on the basis that it can't deal with time-dependent signals like reverb.
 
#116
Looking at the Python, there's a lot of work that can be done to make it much neater and more streamlined but I might well be able to add new training and validation data in. I will have to see how my time looks... Getting the training/validation sets right will be tricky.
 
Last edited:
  • Like
Reactions: meisi and DaBigBadaBoom
#117
Warrenbass01 said:
seconded, the capo capture sounds reallly good.

Another thing I’ve been messing around with is outboard gear captures, NAM is remarkably good at that. Been using a 1073 into a studer a80 capture and it adds this subtle polish and depth of tone that you could really only get with high quality outboard equipment before
Click to expand...
Totally agree with this mindset. It seems there are two camps, accurate as possible and best sound as possible and there is very much a line in the sand between the two.

For me - I’ll take the best sound possible. That would imply on full rig captures running some mastering EQ etc.

Every combination under the sun is fun and all but my ears fatigue and I don’t know what I’m hearing after a bit.
 
  • Like
Reactions: gtirard
#119
meisi said:
Totally agree with this mindset. It seems there are two camps, accurate as possible and best sound as possible and there is very much a line in the sand between the two.

For me - I’ll take the best sound possible. That would imply on full rig captures running some mastering EQ etc.

Every combination under the sun is fun and all but my ears fatigue and I don’t know what I’m hearing after a bit.
Click to expand...
I'm just capturing pedals. Up to the reader to team them with other NAMs for amps and IRs... I know my limits. I still need to test my work so far.
 
  • Like
Reactions: meisi
#120
-Asdfgh- said:
Looking at the Python, there's a lot of work that can be done to make it much neater and more streamlined but I might well be able to add new training and validation data in. I will have to see how my time looks... Getting the training/validation sets right will be tricky.
Click to expand...
Just to expand - the code is a couple of years old and has lots of stuff hard-coded that should really be in configuration files to make it easier to add training inputs. Currently, it checks the training input against overall checksums but also where there are frequency sweeps, latency checks. So one thing to do is to tinker with the code to make it based on configuration files to avoid the need to modify the code to try the next bit.

In terms of training, the first question is will other input wavs work, after adjusting the hashes so it will accept them? If so, what can be done in terms of broadening the things it will take captures of and/or allow tweaking to make it more bass specific? It's quite likely that it won't work, but it's worth a limited investigation.
 

Latest posts

Top Bottom
Back