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.
They describe the datasets used to train their model (licensed data). However, they rely on 10k music tracks that are not described (proprietary data). They do not provide any of the datasets used for training MusicGen. Only a dummy dataset containing just a few examples for illustrative purposes is provided.
Models are available in the GitHub repository and in Hugging Face. Note that the weights are released under the CC-BY-NC 4.0 license as found in the https://github.com/facebookresearch/audiocraft/blob/main/LICENSE_weights.
Training procedure is well documented in the paper, including hardware requirements and model configurations. Additional implementation details are available in the GitHub repository.
Evaluation is documented in the paper and the code provides detailed explanation about the implementation of evaluation metrics. Evaluating the model to reproduce the results still may require some effort. Dataset used for evaluation is MusicCaps benchmark.
The code in this repository is released under the MIT license as found in the https://github.com/facebookresearch/audiocraft/blob/main/LICENSE. The models weights in this repository are released under the CC-BY-NC 4.0 license as found in the https://github.com/facebookresearch/audiocraft/blob/main/LICENSE_weights.
