Indexing Symbolic Music Scores

Indexing non conventional datasets often involves the association of a basic indexing method, used as a building block, with some higher level, application-dependent, matching operations. In the present paper we describe the design of such an integrated index structure for large collections of digital music scores. Our structure requires a textual description of scores content, and relies at its core on well-known n-gram indices. It also encodes in its kernel, with algebraic signatures, some of the complex features that constitute a melodic description. These signatures support, at query time, several variants of pattern-based searches that requires a richer semantics than the mere exact substring matching. As a result, a single instance of our index suffices to cover several interpretations of pattern-based search over large score collections, saving space, time and maintenance efforts. We outline the context, including some introductory explanations on symbolic music management for non-expert readers. We then define the search requirements, and describe in details the structure and its associated operations. We also present experimental results on large music scores collections in a Digital Score Library currently under deployment.