DDSP Guitar

This repository contains demo code for DDSP-based Neural Waveform Synthesis of Polyphonic Guitar Performance from String-wise MIDI Input. The project was conducted together Xin Wang, Erica Cooper, Lauri Juvela, Bob L. T. Sturm and Junichi Yamagishi.

Audio examples here.

Setup

git clone https://github.com/erl-j/ddsp-guitar.git
cd ddsp-guitar
pip install -r requirements.txt

Usage

For rendering a midi file with the unified model, run:

python render_midi.py --midi_path midi.mid --output_path out.wav --crop-seconds 10 --device cuda:0

Use --help for more options.

Acknowledgements

This work is supported by MUSAiC: Music at the Frontiers of Artificial Creativity and Criticism (ERC-2019-COG No. 864189) and JST CREST Grants (JPMJCR18A6 and JPMJCR20D3) and MEXT KAKENHI Grants (21K17775, 21H04906, 21K11951, 22K21319).

DDSP code is based on magenta's ddsp and ddsp_pytorch. Code for dilated convolutions is from @ljuvela.