Indexing and Retrieval of Music Documents through Pattern Analysis and Data Fusion Techniques

One of the challenges of music information retrieval is the automatic extraction of effective content descriptors of music documents, which can be used at indexing and at retrieval time to match queries with documents. In this paper it is proposed to index music documents with frequent musical patterns. A musical pattern is a sequence of features in the score that is repeated at least twice: features can regard perceptually relevant characteristics, such as rhythm, pitch, or both. Data fusion techniques are applied to merge the results obtained using different features. A set of experimental tests has been carried out on retrieval effectiveness, robustness to query errors, and dependency on query length on a collection of Beatles’ songs using a set of queries. The proposed approach gave good results, both using single features and, in particular, merging the rank lists obtained by different features with a data fusion approach.