Design and creation of a large-scale database of structural annotations

This paper describes the design and creation of an unprecedentedly large database of over 2400 structural annotations of nearly 1400 musical recordings. The database is intended to be a test set for algorithms that will be used to analyze a much larger corpus of hundreds of thousands of recordings, as part of the Structural Analysis of Large Amounts of Musical Information (SALAMI) project. This paper describes the design goals of the database and the practical issues that were encountered during its creation. In particular, we discuss the selection of the recordings, the development of an annotation format and procedure that adapts work by Peeters and Deruty [10], and the management and execution of the project. We also summarize some of the properties of the resulting corpus of annotations, including average inter-annotator agreement.