jthickstun / ismir2019coupled

Experiments for Coupled Recurrent Networks for Polyphonic Music Composition

Coupled Recurrent Music Composition

Experiments for Coupled Recurrent Models for Polyphonic Music Composition.

Corresponding author: John Thickstun. University of Washington

---

This repository contains code for the experiments discussed in Coupled Recurrent Models for Polyphonic Music Composition.

Included are notebooks for reproducing the single-part and multi-part experiments presented in Table 3 and Table 4 respectively.

For examples of how to generate output from the model in audio, kern, and midi representations, see sampling.ipynb. A pre-trained model is included for the multipart6 architecture.

Dependencies

• PyTorch (tested with v1.5.0)

• Jupyter (tested with v4.4.0)

• NumPy (tested with v1.18.4)

• SciPy (tested with v1.4.1)

• Mido (tested with v1.2.9)

• Scikit-learn (tested with v0.22.2)

• Matplotlib (tested with v2.1.2)

References

To reference this work, please cite

@article{thickstun2019coupled,
  author    = {Thickstun, John and Harchaoui, Zaid and Foster, Dean P and Kakade, Sham M},
  title     = {Coupled Recurrent Models for Polyphonic Music Composition},
  journal   = {International Society for Music Information Retrieval},
  year      = {2019},
}