Experimenting with Amp Modeling Techniques: Capturing and Training on the Peavey Bandit 112 Silver Stripe

Looking to improve our amp modeling techniques, we conducted experiments modeling the Peavy Bandit 112 Silver Stripe.

Here I share with you the variables of the experimetation, the results, and some interpretations of the results in the end. We focus particularly on the data capture setup.
If you’re just interested in good sounding models of the amp, this is for you.

In the process of capturing the sound of the AMP, we tested 2 “reamping” setups:

  1. Using a DI:


    This setup allowed us to record both the amp-head signal, and the cabinet sound (using a mic) at the same time.

  2. Using a Loadbox:


    Here we only capture the sound of the amp-head.


We captured the sound on 4 different amp configurations:

  1. Clean
  2. Clean2
  3. Crunchy
  4. Gainy

In the training, we used 6 different model architectures

  1. Lightest (LSTM-8) with skip_connection = ON
  2. Light (LSTM-12) with skip_connection = ON
  3. Standard (LSTM-16) with skip_connection = ON
  4. Heavy (LSTM-20) with skip_connection = ON
  5. Light (LSTM-12) with skip_connection = OFF
  6. Standard (LSTM-16) with skip_connection = OFF

RESULTS:

Here’s a directory containing all (60) the resulting models:
https://drive.google.com/drive/folders/1aM9cHo1gyKfGz_Fsu5X9kBuJ0Jd0RHj5

Here’s an excel sheet summarizing the training results of those 60 models (ESR, and final training epochs):


Interpretations:

  1. Comparing setups:


    → This shows that training models on the amp-head signal alone, rather than the sound from the cab, can lead to greater accuracy.
    This is likely due to the simpler signal chain of amp-head alone, compared to the more complex [amp-head → cab] signal chain.
    → After comparing the results of the two setups, we found that setup1, which uses the DI, was more effective at capturing the sound of the amp-head. When we trained models on the amp-head signal using setup1, we noticed that the resulting models are more accurate and (arguably) sounded better to us compared to the models trained using setup2.

  2. Comparing model types:


    → Disabling skip_connection can improve model accuracy. However, this can not be generalized yet, as we have to experiment with more tones and amps.
    → As expected, the bigger the model size, the more accurate they are.

  3. Comparing different amp tones:


    → I cannot derive any conclusions from this. The amp tone doesn’t seem to have an effect on the accuracy of the models.

These intnerpretations may be only valid to this Peavy Bandit amp. In the near future we’re looking to capture more AMPs and pedals!

Please feel free to give feedback on our experiment and how we interpreted the results. If you have any experience with modelling yourself, this can also add more insights!
This is also missing feedback on how the trained models sound, let me know if you tried a bunch of them!

7 Likes

He @itskais ,

I’ve checked out

did you align all tracks and checked for phasing issues?
I might have another understanding of gainy but this has not much gain at all.

Hi @itskais,

What about doing this “reamping” with the moddwarf itself? I mean to play and generate the .wav files, and just use the PC for the “trainer” colab notebook. Would in this case a “reamping box” and a DI or attenuator be needed??

Yoo @spunktsch,

True, the model doesn’t sound as gainy as expected, I believe it’s due to a couple of reasons:

  • Apparently there’s a problem with the input gain with this model, so if you kinda crank up the Pregain knob, and also add pre-gain in the MOD, you get much closer to a gainy sound:

  • Also when capturing the amp, we didn’t go for a very high gain sound because it caused some noise issues (that were due to not using a proper reamp box)

Lemme know if you try this solution

Hi @fer,
I believe using a MOD Dwarf would kinda replace the reamp box and audio interface if you connect it like this:

  • Dwarf output goes to AMP input
  • DI Direct output and MIC go to Dwarf Inputs
  • Usings a File player + Recorder to capture → manually align the files on the computer.

However, this does not cancel the need for a DI or a Loadbox. The output of the AMP is very much high in energy, and the Dwarf cannot get that much energy in the inputs. The DI->CAB or Attenuator are used to attenuate the signal of the AMP so that it’s acceptable by the Audio interface (or Dwarf in this case)

1 Like

Indeed, for Powermp output, you must use a loadbox. (Unless you want to mic the speaker but then you can’t separate amp and speakers)

You could try to manage that with some gain staging if you opt to model only the pre-amp section. (Use the fx send on an amp)

To counter (possible?)latency issues with the input versus target file, I send audio with the audio player to both outputs,let one signal go through a pedal and to input one while the output 2 I just loop back to input 1. Then I take those 2 inputs and record them using a stereo or multi input recorder plugin (by @brummer)

4 Likes

I dig your approach to document your results… thanks for including the well-chosen charts! Looks like they were made with seaborn.

1 Like