MLTerror15

From GitHub - IHorvalds/MLTerror15: Deeply learned simulator for Orange Tiny Terror with Recurrent Neural Networks :tangerine:

File was already in the correct format, I simply edited it to add

  "out_gain": -14,

as otherwise it would be too loud by default.

MLTerror15_0.5-0.5-0.5-model-lstm-1.json (197.2 KB)

Enjoy!

2 Likes

Many thanks, that’s a nice one, but very (too much?) CPU hungry on the Dwarf comparing to others, maybe too deeply learned…? Any possibility to implement a Normal/High fidelity parameter like the Vintage Cabinets plugin ?

1 Like

Ok, I did a re-amp via the the AIDA-X and the Trainer with the original model (73% CPU on the Dwarf)
You’ll find two models here less CPU hungry and usable with cab and reverb, sounds pretty close… ;

MLTerror15_for_Dwarf_Standard 39%
MLTerror15_for_Dwarf_Heavy 48%

https://drive.google.com/drive/folders/1SI-QzCciF5F7GsvO7MfEqlHFvS9mlrkn?usp=share_link

Enjoy!

2 Likes

How is the re-amp done? You don’t have the input, output wav right?

Of course…it’s a kind of virtual re-amp.

  • play the “capture signal” input file through the AIDA-X (loaded with the model you want, EQ bypassed), record both outlets, separate your stereo file and you get your “input.wav” and “target.wav” files → training.

    a bit crude…but it works for me
4 Likes

It was about what I mentioned here

My idea was a unique pedal for transmitting and capturing audio, but using a stereo recorder was a good alternative.

@zwabo, did you had some trouble, or difficulty sync input and output audios?

Hey @jon, I think that this kind of pedalboard could be used as an example on wiki or aidadsp site as an example of capturing heavy plugins for training.

1 Like

Not a problem to sync in Reaper, I did check the samples alignement after separating the L & R channels.

Here is the pedalboard

2 Likes