StemGen: A music generation model that listens
This work addresses the need for music generation models that can listen and adapt to context, offering an incremental improvement over existing methods focused on abstract conditioning.
The authors tackled the problem of generating musical audio that responds to musical context, presenting StemGen, a non-autoregressive transformer-based model that achieves audio quality comparable to state-of-the-art text-conditioned models and demonstrates strong musical coherence.
End-to-end generation of musical audio using deep learning techniques has seen an explosion of activity recently. However, most models concentrate on generating fully mixed music in response to abstract conditioning information. In this work, we present an alternative paradigm for producing music generation models that can listen and respond to musical context. We describe how such a model can be constructed using a non-autoregressive, transformer-based model architecture and present a number of novel architectural and sampling improvements. We train the described architecture on both an open-source and a proprietary dataset. We evaluate the produced models using standard quality metrics and a new approach based on music information retrieval descriptors. The resulting model reaches the audio quality of state-of-the-art text-conditioned models, as well as exhibiting strong musical coherence with its context.