This repository contains the code for the research article:
O. Mikkonen, A. Wright, E. Moliner and V. Välimäki, “Neural Modeling Of Magnetic Tape Recorders,”
in Proceedings of the International Conference on Digital Audio Effects (DAFx),
Copenhagen, Denmark, 4-7 September 2023.
The pre-print of the article can be found from here.
The accompanying web page can be found from here.
The datasets can be found from here.
Clone the repository and submodules
git clone [email protected]:01tot10/neural-tape-modeling.git
cd neural-tape-modeling
git submodule init && git submodule update
Create the Python virtual environment with mamba (preferred) / conda
mamba/conda env create --file environment.yaml
Download data to symlinked location audio/ -> ../neural-tape-audio/
# create a directory for contents
mkdir ../neural-tape-audio
# download and extract toy data
wget -P ../neural-tape-audio 'https://zenodo.org/record/8026272/files/neural-tape-audio_CHOWTAPE.tar'
tar -xzvf ../neural-tape-audio/neural-tape-audio_CHOWTAPE.tar -C ../neural-tape-audio/
# download and extract real data
wget -P ../neural-tape-audio 'https://zenodo.org/record/8026272/files/neural-tape-audio_AKAI.tar'
tar -xzvf ../neural-tape-audio/neural-tape-audio_AKAI.tar -C ../neural-tape-audio/
Optional: To generate target audio with CHOWTape, a VST instance of the plugin should be compiled. Check instructions in the corresponding repository.
The folder scripts/ contains the various processing pipelines for interacting with the system, as well as a separate README.md with instructions.
Cite the work as follows
@conference{mikkonen_neural_2023,
title = {Neural Modeling of Magnetic Tape Recorders},
booktitle = {Proceedings of the {{International Conference}} on {{Digital Audio Effects}} ({{DAFx}})},
author = {Mikkonen, Otto and Wright, Alec and Moliner, Eloi and V{\"a}lim{\"a}ki, Vesa},
year = {2023},
month = sep,
address = {{Copenhagen, Denmark}}
}
🖤🖤🖤
- VST instance of a reel-to-reel tape machine: CHOWTape
- Python VST wrapper: pedalboard
- Dataloader extended and customized from microtcn
- Error-to-signal ratio (ESR) loss from Automated-GuitarAmpModelling
- ESR loss with DC blocker from GreyBoxDRC
- Multi-resolution short-time Fourier transform (STFT) loss from auraloss-library
- Codebase kept clean with with yapf, isort, pylint and beautysh
🖤🖤🖤