Close
0%
0%

NeuralPi

NeuralPi is a guitar pedal using neural networks to emulate real amps and pedals on a Raspberry Pi 4.

Similar projects worth following
NeuralPi is a open source DIY guitar pedal that copies the sound of real amps and pedals. Using models created with machine learning, the NeuralPi can accurately recreate high-end amplifiers or distortion/overdrive pedals. Recordings of the actual amp/pedal are trained by a neural net model, then deployed to the NeuralPi for endless possibilities on your guitar.

The NeuralPi costs around $150 to build, and uses the Raspberry Pi 4b and HiFiBerry DAC + ADC audio card. See links below for a full build guide and demo video. As an open source project, the NeuralPi will continue to evolve and add new features.

NeuralPi was created out of a curiosity for machine learning and a passion for great guitar tone. I had developed several guitar plugins that use machine learning/neural networks, each building on what I had learned from the previous one. I had always had the idea of using this software to build a guitar pedal in the back of my mind. When I started digging into it, I found that much of the groundwork had already been done to make this a reality.  

Using Elk Audio OS, a low latency Linux based operating system, I was able to convert my already developed neural network plugin into a stand-alone guitar effect running on the Raspberry Pi 4. The audio card (HiFiBerry DAC+ADC) was selected out of a desire to get the highest quality sound at the cheapest price. This card provides up to 192kHz/24bit, which is industry standard for high quality digital audio devices. In total, I would only have to spend about $150 to build the NeuralPi, and $50 of that was for the enclosure (optional) and the various audio adapters for using guitar cables with the HiFiBerry input/output.

Let me back up, why would someone want to use neural networks for electric guitar? It turns out that certain neural network models are well suited for the task of emulating the dynamic response of amplifiers and pedals (guitar effects). Musicians pay top dollar for amplifiers with expensive vacuum tube components, because they produce the best sound. Using machine learning, you can take input/output recordings of a particular amp and train a model to behave just like the amp circuitry at a specific setting. This method is effective at modeling tube amps and distortion/overdrive, as well as certain aspects of audio compressors.

  • 1 × Raspberry Pi 4b
  • 1 × HiFiBerry DAC + ADC
  • 1 × Rpi4 Power Adapter (USB-C connector, 5.1V - 3.5A)
  • 1 × Rpi4 + HiFiBerry compatible enclosure (optional)
  • 1 × Dual 1/4" Female to 1/8" Male Stereo Audio Adapter

View all 8 components

  • Next steps

    Keith Bloemer05/28/2021 at 15:09 0 comments

    Completed:

    Milestone 1: Get the neural net plugin running on the Rapsberry Pi with Elk Audio OS, Done!

    To Do:

    Milestone 2: Currently, the NeuralPi plugin running on Elk OS has no user controls. It runs a single model that can be swapped out before running the plugin. Add user controls so that a remote instance of the plugin can control the NeuralPi over Wifi. These controls will include Gain/Volume, EQ, and model selection.

    Milestone 3: Add physical knobs and potentially an LCD screen for selecting models through Elk Audio's Sensei application.

    Milestone 4: While running PyTorch machine learning locally on the Raspberry Pi might be a stretch, it is fully capable of recording high quality audio with the HiFiBerry hat. Implement a capture feature by automating the recording of input/output samples, pushing to remote computer for training, then updating the Pi with the newly trained model.

    Milestone 5: Add traditional effects to use with the neural net models, such as IR (impulse response/cab sim), delay, reverb, chorus, flange, etc. 

View project log

  • 1
    Full Build Instructions

    Click here for full build guide published on Towards Data Science

View all instructions

Enjoy this project?

Share

Discussions

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates