Author Topic: GuitarML Plugins For Pi-Stomp  (Read 848 times)

keyth72

  • Newbie
  • *
  • Posts: 7
    • View Profile
GuitarML Plugins For Pi-Stomp
« on: April 14, 2022, 04:35:16 AM »
I have two LV2 plugins available for Pi-Stomp for anyone interested, an overdrive pedal and an amp. Both of these use machine learning to emulate a specific device.

TS-M1N3: TS-9 Pedal clone
DarkStar: Blackstar HT40 amp clone (gain knob modeled)

The LV2's can be downloaded from the Github page:
https://github.com/GuitarML/modep-plugins

Let me know if you have any issues running them, they are a work in progress. These are not part of the MOD environment yet, so you have to push them to the Pi-Stomp manually, instructions are in the above link.

For more info on what GuitarML is and the kind of plugins I develop, see my website: https://guitarml.com/

Thanks!

-Keith

Randall (Admin)

  • Administrator
  • Full Member
  • *****
  • Posts: 217
    • View Profile
Re: GuitarML Plugins For Pi-Stomp
« Reply #1 on: April 15, 2022, 05:05:40 PM »
This is SUPER Cool!  Machine learned plugins.  That's like the pedal of the future, today!  If it could only learn to make me play better than I really do.

Seriously, thanks a ton for making such an awesome thing, then porting it to pi.

Both of these sound damn good.  A TubeScreamer was an overdrive missing in the arsenal of drives.

I really appreciate that the drive knob on the amp keeps the overall level mostly the same so you don't have to inversely compensate with the volume knob.

Very much looking forward to other ports of your plugins.  Can't wait to hear that Dumble on pi-Stomp!
     


keyth72

  • Newbie
  • *
  • Posts: 7
    • View Profile
Re: GuitarML Plugins For Pi-Stomp
« Reply #2 on: April 18, 2022, 09:19:45 AM »
Thanks Randall! And appreciate all your help along the way, I definitely plan doing more on the PiStomp platform and MOD/MODEP.  And yes, consistent volume seems to be a nice side effect of these neural models, I didn't do anything specific to make that happen.

I made a quick video demo of the plugins, and also updated the release to include the category.
https://youtu.be/WpU084i9w0Q

P.S. You joke about improving guitar playing, but using A.I. to fix "mistakes" and instead play what it thinks you wanted to play is definitely a possibility, if not already been done.  As a guitar player I'm not sure how I feel about that, but the tech is definitely impressive.

Crosenau

  • Newbie
  • *
  • Posts: 9
    • View Profile
Re: GuitarML Plugins For Pi-Stomp
« Reply #3 on: April 19, 2022, 07:35:46 PM »
I loaded them in! amazing!
So stoked you did this.

So far I have the following feedback:
TS :
Input volume bypassed is much higher than output with effect on. I would normally use a TS as boost, this makes that difficult.

Darkstar: Holy crap. its dirty nice. No complaints yet.
Perhaps a low drive model.

Effect suggestions: (dreaming) :-)
Pedals:
Angry Charlie
Charlie Brown

Amps:
Variac Plexi (if you do: max out all the nobs first, use the variac to lower the volume)
Fender Deluxe Reverb (bright and normal)

keyth72

  • Newbie
  • *
  • Posts: 7
    • View Profile
Re: GuitarML Plugins For Pi-Stomp
« Reply #4 on: April 21, 2022, 10:38:22 AM »
Thanks for the feedback! Good observation on the TS, should probably add a general boost to make it act more like the pedal.

I'm planning on adding more models to Darkstar in the next release, basically just adding a dropdown menu to switch amp models. I want to keep it as simple as possible and not make it bogged down with generic EQ or other effects, since there's already so much available.

As for which amps/pedals to add, I'll be relying on the community for most of these. I have a couple of sources already but if there's anyone reading this that wants to contribute, you can email me at "smartguitarml@gmail.com" and we can go from there. Basically all I need are audio recordings from the amp/pedal, where a specific input signal is played through the device and the output is recorded. I can provide the input signal in the form of a WAV file. If you have an audio interface, it's pretty easy to play the input WAV through your device and record the output simultaneously, either by direct out or from a microphone.

I'll be adding single parameter models to Darkstar, typically a gain/drive. These models require 5 separate recordings, so one 3-minute recording for each 0%, 25%, 50%, 75%, and 100% of a gain knob.  I can then use these to train a neural network model that *hopefully* sounds close to the real thing.

Edit: I should note that this modelling process works for distortion/overdrive pedals, clean to overdriven amps, but does not capture reverb/delay/flange/phaser, etc. I have had good results with saturation effects, and mixed results with compressors.
« Last Edit: April 21, 2022, 10:52:09 AM by keyth72 »