GRAFX

GRAFX is an open-source library designed for handling audio processing graphs in PyTorch. One can create and modify a graph, convert it to tensor representations, and process output audio efficiently in GPU with batched node processing. The library is complemented with various differentiable audio processors, which enables end-to-end optimization of processor parameters or their estimators (e.g., graph neural networks) via gradient descent. The code can be found in this github repository.

Installation

pip install grafx

Some processors use convolutions; for their efficient processing, install FlashFFTConv from the following github repository.

Contents

Introduction

• Audio Processing Graphs
• Differentiable Processors
• Batched Audio Processing

Graph API

• grafx.data
• grafx.render
• grafx.draw
• grafx.utils

Processor API

• grafx.processors.core
• grafx.processors.filter
• grafx.processors.eq
• grafx.processors.stereo
• grafx.processors.dynamics
• grafx.processors.nonlinear
• grafx.processors.reverb
• grafx.processors.delay
• grafx.processors.container

References

• Versions
• References

Citation

@inproceedings{lee2024grafx,
title={{GRAFX}: an open-source library for audio processing graphs in {P}y{T}orch},
author={Lee, Sungho and Mart{\"\i}nez-Ram{\"\i}rez, Marco A and Liao, Wei-Hsiang and Uhlich, Stefan and Fabbro, Giorgio and Lee, Kyogu and Mitsufuji, Yuki},
booktitle={DAFx (Demo)},
year={2024}
}