Music-Generation

This is a toolkit for symbolic music generation in pyTorch. You can generate tracks from scratch (unconditional) or continuation for a given track.

Features

• Using pytorch_lightning as backend.
• Using Deepmusic for encoding/decoding MIDI files.
• Supporting different transformer models + vanilla rnn. (vanilla, linear, huggingface transformers)
• Supporting Compound Word and REMI representations. (see references)

Usage

Please see main.ipynb to learn how to train a model and generate music!

Generated Samples

check out assets directory!

References

[1] https://github.com/YatingMusic/compound-word-transformer

[2] https://github.com/YatingMusic/remi

[3] https://github.com/idiap/fast-transformers

[4] https://github.com/huggingface/transformers