This is the repository for the paper 'Open-Amp: Synthetic Data Framework for Audio Effect Foundation Models'

Open-Amp is a framework for creating large scale and diverse audio effects data using crowd-sourced audio effects models. This repository shows basic usage of the package for training a universal amplifier model, as described in the paper.

<div style="text-align: center"> <a href="https://alec-wright.github.io/OpenAmp/" class="btn btn--primary btn--small" target="_blank" rel="noopener noreferrer"> 🔊 Audio Examples </a> </div>

Download amp models and clean guitar data

Use the following code to download the 'Proteus Tone Packs' collecton of amplifier and effect pedal models.

curl -LO https://github.com/GuitarML/ToneLibrary/releases/download/v1.0/Proteus_Tone_Packs.zip
tar -xf Proteus_Tone_Packs.zip
rm Proteus_Tone_Packs.zip

Use the following code to download clean input guitar for use during training from the IDMT-SMT-GUITAR dataset

curl -LO https://zenodo.org/records/7544110/files/IDMT-SMT-GUITAR_V2.zip
tar -xf IDMT-SMT-GUITAR_V2.zip
rm IDMT-SMT-GUITAR_V2.zip

Environment

install the conda environment from the environment.yaml (you must have the conda package manager installed already)

conda env create -f environment.yaml
conda activate open-amp-demo

Compile input audio files

This goes through the downloaded clean guitar audio, trims silence and produces a single wav file for each of the guitars, in the 'Data' directory

python compile_input_data.py -o 'Data/Ibanez2820-DI' -i 'IDMT-SMT-GUITAR_V2/dataset4/Ibanez 2820'
python compile_input_data.py -o 'Data/CareerSG-DI' -i 'IDMT-SMT-GUITAR_V2/dataset4/Career SG'

Run Universal Amp Training

This runs a basic version of training the universal amp model

python train_universal_amp.py

Use Pre-Trained Amp Model

This is an example of loading a pre-trained universal amp model, trained using OpenAmp, from a checkpoint and running inference on some input guitar audio

python run_uni_amp.py

Use Pre-Trained Fx-Encoder

This is an example of loading a pre-trained Fx-Encoder, trained using OpenAmp, from a checkpoint and extracting encodings from data

python run_fx_enc.py