Universal Optical Music Recognition

Optical music recognition works well on clean, engraved scores — and falls apart on everything else: handwritten manuscripts, dense orchestral pages, non-standard notation. This project builds a universal OMR framework that holds up on visually complex notations, and explores real-time human–GenAI interaction for controllable music generation on top of it.

why it matters

Most music GenAI systems treat scores as an afterthought. Making notation machine-readable at scale unlocks controllable generation: a musician edits the score, the model responds in real time.

status

Active — building the recognition framework and prototyping the interaction loop.