Jump to content

SmartAmpPro


marlowg01

Recommended Posts

This machine learning plugin is amazing.  It is CPU heavy, but the tone is fantastic.

Guitar plugin made with JUCE that uses neural network models to emulate real world hardware. This plugin uses a LSTM model to recreate the sound of real amps and pedals. You can record samples and train models from the plugin. Tone models are saved in .json format. Model training is accomplished using Tensorflow/Keras. The main improvement from the original SmartAmp is that training takes less than five minutes on CPU (vs. 8 hours on GPU) for comparable sound quality. Training has also been integrated into the plugin.

https://github.com/GuitarML/SmartAmpPro

  • Thanks 1
Link to comment
Share on other sites

I just gave it a test drive and, IMHO, it sounds stellar and is well worth a look.  For a freebie it's a must-have.  FWIW I didn't find it hit my machine that hard.  Also, for my DAW (Studio One) to find it I had to install it in a VST3 specific folder rather than a general VST folder.  

Edited by jude77
Link to comment
Share on other sites

If anyone wants me to convert a model for them, I can.  Just send me the direct guitar recording and the one from the plugin or hardware pedal or amplifier and I can convert and post or email the .JSON file for you.  The recording has to be about 3.5min long to get the best model.  The plugin was struggling to work for me for conversions, so I am doing it via command line syntax to create the models.  I could also share how to do that for others if there is interest.

 

Here are a couple of other models that I think turned out well, too, taken from a plugin and from the G1X4 pedal from Zoom:

 

G1X4_DZ_DRV.json G1X4_ORG120.json G1X4_NYC_Muff.json Marshall_Super_Lead_1959_Plexi_100W.json

  • Thanks 4
Link to comment
Share on other sites

It was unusable here - as soon as it was inserted into an FX bin, CbB's GUI ground to a halt and a couple of times, I had to Task Manager kill CbB to close it.  My PC isn't cutting edge, but it's no slouch (Ryzen 5, 6/12 core, 3.4GHz, 32GB RAM, nVidia 1030 graphics card).

Link to comment
Share on other sites

9 hours ago, marlowg01 said:

The plugin was struggling to work for me for conversions, so I am doing it via command line syntax to create the models.  I could also share how to do that for others if there is interest.

I would be interested in learning how to do this.  I downloaded the software but am completely baffled by it.  Are there instructions anywhere?

Link to comment
Share on other sites

So I actually started working with these machine learning plugins with the PedalNet, where you download the git package and open the documents for setting up the software.  This video describes the software setup and then you can skip and download the SmartGuitarPro part and use the commandline at the bottom of this post.

Once I had completed that, I also installed the pip requirements for the SmartGuitarPro software.  However, once I ran the plugin, a commandline window opened that said that I needed to install python and tensorflow, etc.  Instead, I ran the terminal through Anaconda3 which I had used for the previous plugin and ran the following command lines for each stereo, hard-panned, un-effected on the left, effected on the right:

(base) C:\Users\<username>\AppData\Roaming\GuitarML\SmartAmpPro\training>python train.py C:\Users\<username>\<location_of_stereo_wav_file(uneffected-Left, effected-Right)>\G1X4_MS800.wav G1X4_MS800

I kept the name of the wave file, but removed the *.wav extension and that is the name of the model saved.  It saves in the models and tones folders in this location: C:\Users\<username>\AppData\Roaming\GuitarML\SmartAmpPro\

When you reopen the plugin, it re-scans the *.JSON models and you can access them from the pull-down.  

Each model takes about 5 minutes because it splits the data and using the power of your CPU to process the data more quickly than the GPU-accelerated PedalNet variant, which takes 8 hours to 2.5 days to create models depending upon your computer and hardware.  I downloaded the non-cuda pytorch package because I do not have any GPUs that support either machine learning algorithms.   

I have spoken to the developer about my CPU woes with this plugin as well, and he suggested compiling for the specific system that you have to improve performance.  I plan to do this and look at the code to see if I can see any optimizations that can reduce CPU.  I have found that closing the GUI greatly reduces CPU, so I think that the GUI coding is probably trying to update too frequently, but I am not sure.

Post what you create if you can!  I'd love to see what others produce too.  And here are a few new ones I made that I think also turned out pretty well:

 

 

G1X4_HG_Throttle.json G1X4_RedCrunch.json G1X4_UK30A.json ENGL_Savage_120.json

  • Thanks 2
Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...