Fast Index Based Filters for Music Retrieval

Abstract

We consider two content-based music retrieval problems where the music is modeled as sets of points in the Euclidean plane, formed by the (on-set time, pitch) pairs. We introduce fast filtering methods based on indexing the underlying database. The filters run in a sublinear time in the length of the database, and they are lossless if a quadratic space may be used. By taking into account the application, the search space can be narrowed down, obtaining practically lossless filters using linear size index structures. For the checking phase, which dominates the overall running time, we exploit previously designed algorithms suitable for local checking. In our experiments on a music database, our best filter-based methods performed several orders of a magnitude faster than previous solutions.