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.
Cascaded text-to-music generation model based on conditional flow matching, capable of zero-shot infilling and continuation.
Application types:Text-to-music
Architecture:Flow matching
Essential categories
Category
Status
Notes
Source code
✘ Closed
Despite the arXiv version of the paper stating "Our code and model will be publicly available", no official source repository, training framework, or inference code has been released.
Training data
✘ Closed
MusicFlow was trained on 20K hours of proprietary music. No dataset access, tracklists, origin metadata, or permissions are disclosed.
The research paper details the mathematical formulations of Conditional Flow Matching, loss objectives, two-stage cascaded flow architecture (semantic and acoustic networks), and core hyperparameters. However, hardware requirements and specific model configurations are missing.
Evaluation protocols are thoroughly documented using the public MusicCaps benchmark. It includes standard objective metrics (FAD, FD, KL divergence, CLAP text-audio similarity) and human listening evaluations for text-to-music, infilling, and continuation.
An official supplementary website showcases audio generation examples across text-to-music, continuation, and infilling tasks, along with comparisons against AudioLDM.
