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.

DiffRhythm

ASLP@NPU and Shenzhen Research Institute of Big Data · 2025

Song generation model able to synthesize complete songs with both vocal and accompaniment up to 4m45s.

Application types: Text-to-songLyrics-to-song

Architecture:Latent diffusion

Essential categories

Category	Status	Notes
Source code	✔︎ Open	Souce code is avaiable in the GitHub repo, including model architecture, inference, training and dataset processing.
Training data	✘ Closed	Training data is not available nor properly described. In the research paper, they indicate that the model was trained on 60,000 hours of audio content, including English (60%), Chinese (30%) and instrumental (10%) songs (about 1 million songs). However, no information on specific data sources, acquisition methods, and data licensing is given. See "Dataset Setup" in Section 4.1. of the pre-print.
Model weights	✔︎ Open	Model weights are avaiable in their HuggingFace page. Multiple versions of the model are avaiable.
Code documentation	~ Partial	Code documentation inclues instructions on how to install and inference the model, but there are no instuctions on how to train it (it is indicated at "coming soon").
Training procedure	✔︎ Open	Training procedure is properly described in the research paper, including details on the models parameters and hardware requirements.
Evaluation procedure	~ Partial	Evaluation procedure is described in the research paper, including metrics (see 4.3) and detailed results (see 5). However, evaluation data is not sufficently described nor avaiable to reproduce the evaluation.
Research paper	~ Partial	Research paper is available as preprint in arXiv. Also for more recent versions of the model: DiffRhythm+ and DiffRhythm 2. No peer reviewed version is avaiable for any of the papers.
Licensing	✔︎ Open	Apache 2.0 License

Desirable categories

Category	Status	Notes
Model card	⭐ Included	Model card available in Hugging Face.
Datasheet	∅ Not included
Package	∅ Not included
User-oriented application	⭐ Included	A demo is available in HuggingFace.
Supplementary material page	⭐ Included	A supplementary page is available, including a brief description of the model, links to relevant sites (e.g., demo, paper) and some examples showcasing the model's abilities.

Raw YAML file with complete evaluation.