Proteus Pedal - Using Ai to Mimic Real Amps and Pedals

Community Elk Showcase
1

Hey! Wanted to show off my latest creation that uses Elk OS on a RaspberryPi. I took my previous NeuralPi project and turned it into an actual guitar pedal, and made some major improvements to the neural models. I’m planning on continuing to build on and improve it.

Thanks Elk team for making this possible!

Here’s my YouTube video featuring sound demos and hardware breakdown:

4 Likes
2

Wow, really impressive and thanks for sharing. Nice that you also added a remote gui to it.

3

Thanks for sharing @keyth72 , really awesome stuff!

4

Amazing!
In a similar space is the aida-x technology, I wonder how it compares especially wrt CPU load?
S’manth x

1 Like
5

Yes! It is very similar, both Proteus and Aida-X use the Automated-GuitarAmpModeling code for training and RTNeural for real time processing. On Aida x they are using a smaller neural net to run efficiently on their hardware and in the Mod environment (LSTM size 16 by default) and I’m running an LSTM size 40 in my Proteus plugin. I was seeing about 50% cpu usage on the Rpi4 with the size 40 models, so pretty significant cpu usage. I was previously running size 20 with around 20% usage.

1 Like
6

Cool.
Does a higher LSTM results in higher quality sound?
Is the LSTM a runtime setting or does it affect the training/model?
I wonder if the LSTM could be a LV2 parameter that the user could select?
S’manth x