MusGO Framework: Assessing Openness in Music-Generative AI

This website builds on the paper MusGO: A Community-Driven Framework for Assessing Openness in Music-Generative AI, authored by Roser Batlle-Roca, Laura Ibáñez-Martínez, Xavier Serra, Emilia Gómez, and Martín Rocamora, and published in the Proceedings of the 26th International Society for Music Information Retrieval Conference (ISMIR 2025). It serves not only as a companion to the publication, but also as a living resource, which is continuously updated and shaped by contributions from the community.

DDSP

Google Magenta · 2020

A library that enables direct integration of classic signal processing elements with DL models.

Application types: MIDI-to-audioAudio synthesisStyle transfer

Architecture:Differentiable DSP

Essential categories

Category	Status	Notes
Source code	✔︎ Open	The complete source code is publicly available on GitHub, including core differentiable modules, training pipeline and inference code.
Training data	✔︎ Open	The research paper indicates that the autoencoder is trained on public, fully open-source datasets including NSynth and individual instrument stems from the URMP dataset (see Section 4.2). The repository includes open scripts (`ddsp_prepare_tfrecord`) to easily format custom training data.
Model weights	~ Partial	Pre-trained model weights are available in demos and tutorial Colab notebooks, and can be downloaded from there. However, no independent downloadable weights are provided.
Code documentation	✔︎ Open	The codebase is well documented, with a detailed README file and additional documentation for the core differentiable modules. The repository also includes example notebooks for training and inference, providing clear guidance on how to use the code.
Training procedure	✔︎ Open	The training procedure is described in the research paper. It is also well-documented in the GitHub repository, with clear instructions and scripts provided for training the model.
Evaluation procedure	✔︎ Open	The evaluation procedure is described in the research paper (see Appendix C, for further details).
Research paper	✔︎ Open	Accepted at ICLR 2020.
Licensing	✔︎ Open	The project is licensed under the Apache License 2.0.

Desirable categories

Category	Status	Notes
Model card	∅ Not included
Datasheet	∅ Not included
Package	⭐ Included	It is distributed as an indexed, versioned Python software package that can be installed via pip (`pip install ddsp`).
User-oriented application	⭐ Included	DSSP is a highly accessible model to end-users and creators through user-friendly interactive Google Colab notebooks, web-based demos, and an official real-time VST3/AU audio plugin (ddsp-vst) for digital audio workstations (DAWs).
Supplementary material page	⭐ Included	There is a dedicated blog and supplementary page featuring interactive listening examples, sonified timbre transfers, and explanations of the core architecture.

Raw YAML file with complete evaluation.