What it is
MIDI2Score explores whether symbolic music transcription can be framed as a sequence-to-sequence problem, with MIDI token sequences as input and MusicXML token sequences as output.
Problem
Turning raw symbolic music data into readable score structure is not just a formatting task. It requires a representation that can preserve musical intent while staying learnable for a sequence model.
Approach
The project uses a Transformer-based setup and treats the conversion problem as token-to-token translation. That keeps the framing close to modern sequence modeling workflows while staying focused on symbolic music rather than audio.
Why it matters
Even in its early state, the project is a useful experiment at the intersection of generative modeling, structured outputs, and music representation.