Projective Orchestration Dataset (POD)

A MIDI database for musical arrangement

The Projective Orchestration Dataset (POD) is a MIDI collection of 180 pairs of piano scores and corresponding orchestration performed by famous composers. Its purpose is to offer a framework for studying the relations between a piano score and its orchestral arrangements from a symbolic point of view.

It is organized in 180 folders containing :

  • a midi piano score
  • a corresponding midi orchestral score 
  • a coma-separated value (CSV) file linking the midi track names of the orchestral file to standardized instrument names (for example associating the track name vn1 to the instrument violin).

In our opinion, a particularly interesting problem that could benefit from this database is to observe and possibly infer the transformation from a piano score to an orchestral version (orchestration) or its inverse transformation (reduction). To tackle this task, temporally aligned versions of piano scores and there orchestral counterparts are necessary. Hence, we also provide an automatic alignment tool relying on Needleman-Wunsch algorithm and that can be found, along with the database, on the webpage of the project :

Course Syllabus