This page contains additional materials for the paper
L. Angioloni, T. Borghuis, L. Brusci, and P. Frasconi. CONLON: A Pseudo-Song Generator Based on a New Pianoroll, Wasserstein Autoencoders, and Optimal Interpolations. In Proceedings of the 21st International Society for Music Information Retrieval Conference, ISMIR 2020.
CONLON is a pattern-based MIDI music generator with three main features:
- A novel pianoroll-like description of patterns
- Wasserstein eutoencoders as the underlying generative model
- Optimal trajectories for interpolation and swirls.
CONLON was developed as a collaborative effort between the AI Lab at DINFO, University of Florence, (Italy) and MUSI-CO, a music generation startup based in Eindhoven (the Netherlands).
Source code
We are sorry that source code cannot be provided at this time (please do not ask, if code can be made available it will appear here). The WAE implementation of Ilya Tolstikhin is a good starting point for developing the generative model.
Acknowledgments
We would like to thank Fabio Schoen for useful discussions. We are grateful to Volker Böhm (Musik Akademie Basel/Fachhochschule Nordwestschweiz), Luigi Ceccarelli (Conservatorio di Latina), Marek Choloniewski (SME studio, Music Academy of Krakow, University of Krakow), Josef Gruendler (FH Joanneum, Graz), Michał Janocha (SMEAMuz, Poznań), Marco Ligabue (Academia di Belle Arti di Firenze), Adam Stanović (University of Sheffield), and Alla Zagaykevych (National Music Academy of Ukraine) for they valuable help in recruiting musicians who were involved in the listening surveys.