Alberto Pinto

A novel XML music information retrieval method using graph invariants

The increasing diffusion of XML languages for the encoding of domain-specific multimedia information raises the need for new information retrieval models that can fully exploit structural information. An XML language specifically designed for music like MX allows queries to be made directly on the thematic material. The main advantage of such a system is that it can handle symbolic, notational, and audio objects at the same time through a multilayered structure. On the model side, common music information retrieval methods do not take into account the inner structure of melodic themes and the metric relationships between notes. In this article we deal with two main topics: a novel architecture based on a new XML language for music and a new model of melodic themes based on graph theory. This model takes advantage of particular graph invariants that can be linked to melodic themes as metadata in order to characterize all their possible modifications through specific transformations and that can be exploited in filtering algorithms. We provide a similarity function and show through an evaluation stage how it improves existing methods, particularly in the case of same-structured themes.

A generalized graph-spectral approach to melodic modeling and retrieval

The development of both musicologically based and efficient music information retrieval metrics to query large music database is crucial in modern music information retrieval, knowledge management and database research. Graph spectral representation of pitch class sequences has proved to outperform other pitch class based melodic similarity methods. Here we compare different spectral approaches to structural queries in databases of symbolic music, which exploits mathematical music theory results to improve the descriptive power of representative graphs. In particular, we explore graph representation of other relevant music features like intervals. The experiments have been conducted on a subset of the RISM collection, and results have been evaluated against a ground truth for the same collection developed for the MIREX competition.

Multi-model Music Content Description and Retrieval Using IEEE 1599 XML Standard

The new format IEEE 1599 for music and audio content description defines a standard for the representation of retrieval models within music and music/audio formats that makes use of XML documents as content descriptors. In this article, it is described how music/audio semantics can be actually represented within the Structural Layer of IEEE 1599, thanks to the introduction of novel Music Information Retrieval (MIR) objects that can be exploited by music search engines. A complete description of MIR objects is provided and it is shown how they can be used to embed metadata relative to specific music retrieval models, thus allowing for the description of music content in different retrieval contexts. To this aim, a new concept for MIR Model is introduced together with its formalization and tools provided by category theory. The role of MIR objects and morphisms in music content description and retrieval is explained. Furthermore, a concrete example is given with the implementation of a graph-based model within the IEEE 1599 framework.

A graph theoretic approach to melodic similarity

Common music information retrieval methods are based upon editing distances, reductionism or functional analysis tecniques. We adopt an approach which looks into a thematic fragment (TF) globally. This leads to associate a musical graph to each TF which preserves its more abstract content. Then, necessary conditions for graph inclusion are introduced and we give a similarity function between graphs which allows to assign different weights to the elements belonging to different graph powers. The advantage is that graphs catch more musical transformations than other methods, like permutations of subfragments.

An IEEE 1599-Based Interface for Score Analysis

This paper deals with a software application to visualize score analysis. Such a tool allows both the enjoyment of music in general, thanks to the synchronized execution of heterogeneous multimedia objects, and the highlighting of the results of score analyses. As a consequence, the user can investigate the compositional process and the relationships among the music objects of a given piece while listening to an audio track and viewing the corresponding score. This is made possible by a new XML-based standard for music representation --- namely IEEE 1599 --- which will be shortly described.

Method and Environment for the Retrieval of Structural Music Information

Existing solutions for searching for appropriate music content rely on methods like string edit distances or functional approaches (as for the symbolic level) or features extraction methods (as for the audio level) that cannot provide structural similarity information. Our results in structural similarity measures for music information retrieval rely on graph theory and provide new tools for extract characterizing structural quantities from musical data (both symbolic and audio) to be implemented within an integrated query environment. The improvements regard the use of the graph model of musical data at different levels, the MX format for the representation of musical work and the new user interfaces for the navigation of musical source material

Relational motif discovery via graph spectral ranking

Music summarization aims at finding the most representative parts of a music piece (motifs) that can be exploited for efficient music indexing. Here we present a novel approach for motif discovery in music pieces based on an graph spectral ranking. Scores are segmented into a network graph of music segments and then ranked depending on their centrality. Different poli- and mono-phonic metric concepts can be adopted to compare music segments. Bars with higher centrality are more relevant for music summarization. We present an evaluation on the corpus of J. S. Bachs 2-part Inventions both in poli- and mono-phonic configuration.

A Framework for Music Content Description and Retrieval

The recently approved format for music content description IEEE PAR1599 (MX) defines a standard for retrieval models representation within music and audio/video formats that makes use of XML documents as content descriptors. We show how music/audio semantics can be represented within the Structural layer of MX through the introduction of novel Music Information Retrieval (MIR) objects in order to embed metadata relative to specific retrieval models.