MusicExtender

This is the official implementation of MusicExtender, which is a phrase-level autoregressive music generation model based on piano roll music representations and diffusion models, capable of generating polyphonic piano music with long-term structure and controllable phrase length.

Here are some generated musical examples:

Example 1    Example 2    Example 3    Example 4

Example 5    Example 6    Example 7    Example 8

Example 9    Example 10    Example 11    Example 12

Acknowledgments

We appreciate Polyffusion for providing the reference codes.