Scripts, Files, and Resources for Constructing a Large-scale Dataset of Blackbox Effects for Timbre Transfer
-
Knoblin is a (somewhat) user-friendly repackaging of the servo controller aspect of the Amp-Space data collection, used for sampling real devices (guitar pedals, amplifiers). It has since been updated with a GUI and some improved usability features to help reproduce the configuration used in the Amp-Space paper, across different devices, with different 3D printers, servos, and respective tolerances.
-
Tone Render is a package for generating audio samples from VSTs, performing parameter sweeps in a way analogous to how Knoblin sweeps physical knobs. Any VST which can be loaded into REAPER should be compatible. It also includes a set of DI files (both real and synthetic) and scripts to generate scales / note DIs.
-
Amp-Space Dataset
The Amp-Space Dataset was rebuilt from the ground-up to be more diverse, higher-quality, and user friendly (and to address reviewer concerns) than the originally proposed version. For distribution, it is currently hosted in separate files, and made available through a HuggingFace Datasets style loader:
https://huggingface.co/datasets/narad/rockingface
Loading the dataset is as easy as:
from datasets import load_dataset
dataset = load_dataset("narad/rockingface")
It is currently just a test release with a handful of devices, but will be adding more data files once the interface is tested.
The code and data in this repository was originally developed for and described in Amp-Space, an on-going project in machine learning and interface design for music, focused primarily on timbre-changing effects for guitar:
@article{narad2021ampspace,
title = {Amp-Space: A Large-scale Dataset for Fine-grained Timbre Transformation},
author = {Naradowsky, Jason},
journal = {Proceedings of the 24th International Conference on Digital Audio Effects (DAFx-21), Vienna, Austria},
year = {2021}
}