The past four months I’ve been focused on my fellows project at OpenAI: MuseNet. It’s a music generation model (inspired by my music project last summer, but with a much larger transformer architecture, and a significantly larger dataset). We recently published a blog post about it, where you can try co-composing a piece with the model.
I’m very excited to continue working with musicians (and non-musicians!) who want to compose using this model. Please feel free to contact me with suggestions for the model, or particularly if you have midi files or new genres to add to the dataset.
I’ll also be publishing a fuller description the model soon (as well as publishing the code and the model weights).
In the meantime, you can listen to a podcast of me describing MuseNet here.
