discovered 03 Aug 2026
musegan
→ View on GitHubMuseGAN is a music generation tool designed to create polyphonic compositions across multiple instrument tracks, either from scratch or as accompaniment to a user-provided track. It leverages neural network architectures, particularly BinaryMuseGAN, utilizing 3D convolutional layers to handle temporal music structure, allowing for the generation of complex musical phrases influenced by the Lakh Pianoroll Dataset. Notable features include support for training with custom data, the ability to set up and run experiments, and the option to use pretrained models for inference.