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.

MG²

Guangxi University, China and Southwestern University of Finance and Economics, China · 2024

A lightweight melody-guided text-to-music model. Melody information is incorporated implicitly within the CLMP and explicitly in the retrieval augmented diffusion module.

Application types: Text-to-musicMelody-to-music

Architecture:Diffusion

Essential categories

Category	Status	Notes
Source code	✔︎ Open	Source code is available on GitHub, including data processing, model architecture, training and fine-tuning, and inference scripts.
Training data	✔︎ Open	This model was trained on about 132 hours from two open datasets: MusicCaps and MusicBench, which are detailed in Hugging Face.
Model weights	✔︎ Open	Checkpoints are available on Hugging Face.
Code documentation	✔︎ Open	Source code is well-documented, including environment setup,data processing, training and fine-tuning procedure, as well as quick start instructions.
Training procedure	✔︎ Open	Training procedure is fully described in the research paper (see Section 5.3). Code documentation provides additional instructions on how the model might be retrained or fine-tuned.
Evaluation procedure	✔︎ Open	Evaluation procedure is fully described in the research paper (see Section 5.4). Evaluation data is 1/10 of MusicCaps and MusicBench. Testing splits are properly identified in the dataset card. In addition, a human evaluation was conducted to assess recognizability, text relevance, satisfaction, quality and market potential of generated samples (see Section 6). These samples are shared in their complementary website.
Research paper	~ Partial	Research paper is available on arXiv, but it has not yet been peer-reviewed.
Licensing	✔︎ Open	The project is licensed under the MIT License.

Desirable categories

Category	Status	Notes
Model card	⭐ Included	A model card is available on Hugging Face.
Datasheet	⭐ Included	A dataset card is provided on Hugging Face.
Package	∅ Not included
User-oriented application	∅ Not included	An online-based user-oriented application should be available in the indicated link. However, it is not currently accessible.
Supplementary material page	⭐ Included	An official project website showcases sonified audio samples generated by MG² alongside comparative baselines (AudioLDM2, Mustango).

Raw YAML file with complete evaluation.