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.

Magenta RealTime

Google Magenta · 2025

High-quality real-time music synthesis with low-latency real-time controls via MIDI, text, and audio.

Application types: MIDI-to-audioText-to-music

Architecture:Autoregressive transformer

Essential categories

Category	Status	Notes
Source code	~ Partial	The codebase is available on GitHub, including model architecture, and inference code. However, the training code is not publicly available.
Training data	✘ Closed	Technical report indicates the model has been trained on ~190,000 hours of instrumental stock music. But there is no clear description of the datasets or specifc sources used for training.
Model weights	✔︎ Open	Model weights are available on Hugging Face.
Code documentation	✔︎ Open	Model code is well documented in the GitHub repository, including installation requirements, usage instructions, and examples.
Training procedure	✔︎ Open	Training procedure is documented in the technical report, including hardware requirements and model configurations.
Evaluation procedure	✔︎ Open	Evaluation procedure is documented in the technical report, including evaluation metrics and results. Magenta Realtime is compared against two other music genrative models (Stable Audio Open and MusicGen) using the Song Describer dataset.
Research paper	~ Partial	A comprehensive technical report is available. No peer-reviewed publication has been published yet.
Licensing	✔︎ Open	The code in this repository is released under Apache 2.0 License. The models weights in this repository are released under the CC-BY-NC 4.0 license.

Desirable categories

Category	Status	Notes
Model card	⭐ Included	Model card is available on Hugging Face.
Datasheet	∅ Not included
Package	⭐ Included	Magenta Realtime is available as a Python package, installable via PyPI (`magenta-rt`).
User-oriented application	⭐ Included	There is a demo notebook available on Google Colab, which allows users to interact with the model in real-time.
Supplementary material page	⭐ Included	There is a dedicated blog and supplementary page featuring listening examples, demonstrations, and explanations of the core architecture.

Raw YAML file with complete evaluation.