Facebook's MusicGen Allows for Simple and Controllable Music Generation with a Single Language Model

Facebook has introduced MusicGen, a single Language Model (LM) that can generate high-quality music samples, while being conditioned on textual description or melodic features. With efficient token interleaving patterns, MusicGen eliminates the need for cascading several models, and is superior to the evaluated baselines on a standard text-to-music benchmark. Music samples can be found on the supplemental materials, and code and models are available on their GitHub repo. Through ablation studies, Facebook has shed light on the importance of each of the components comprising MusicGen, making it a promising tool for music producers.