This work is still under development
Torch implementation for an auto-regressive sequential generative model that can learn from a single multi-track music segment, to generate coherent, aesthetic, and variable polyphonic music of multi-instruments with an arbitrary length of bar. Such as:
- JSB Chorale datasets pickle or MIDI.
- some multi-track MIDIs.
Put MIDIs in train_data/, such as train_data/JSB-Chorales-dataset/*.pickle
and train_data/midi/*.mid
- train and evaluate with MIDI
make train CUDA=0 NAME=test TYPE=xl N=400 DIR=midi FILE=xxx.mid
make test CUDA=0 NAME=test TYPE=xl DIR=midi FILE=xxx.mid
- train and evaluate with pickle (JSB chorales dataset)
make train_pickle CUDA=0 NAME=test TYPE=xl N=400 DIR=JSB-Chorales-dataset FILE=jsb-chorales-16th.pkl INDEX=0
make test_pickle CUDA=0 NAME=test TYPE=xl DIR=JSB-Chorales-dataset FILE=jsb-chorales-16th.pkl INDEX=0
If you use this code for your research, please cite our paper:
@inproceedings{sintra2021,
title={Sin{T}ra: Learning an inspiration model from a single multi-track music segment},
author={Qingwei Song, Qiwei Sun, Dongsheng Guo, Haiyong Zheng},
booktitle={Proceedings of the 22nd Conference of the International Society for Music Information Retrieval},
year={2021}
}
Feel free to contact me! songqingyu@stu.ouc.edu.cn