Music Retrieval: A Tutorial and Review

Abstract

The increasing availability of music in digital format needs to be matched by the development of tools for music accessing, filtering, classification, and retrieval. The research area of Music Information Retrieval (MIR) covers many of these aspects. The aim of this paper is to present an overview of this vast and new field. A number of issues, which are peculiar to the music language, are described--including forms, formats, and dimensions of music--together with the typologies of users and their information needs. To fulfil these needs a number of approaches are discussed, from direct search to information filtering and clustering of music documents. An overview of the techniques for music processing, which are commonly exploited in many approaches, is also presented. Evaluation and comparisons of the approaches on a common benchmark are other important issues. To this end, a description of the initial efforts and evaluation campaigns for MIR is provided

Indexing music by mood: design and integration of an automatic content-based annotator

“…In the past few years, research in content-based techniques has been trying to solve the problem of tedious and time-consuming human indexing of audiovisual data. In particular, Music Information Retrieval (MIR) has been very active in a wide variety of topics such as automatic transcription or genre classification [5,29,41]. Recently, classification of music mood has become a matter of interest, mainly because of the close relationship between music and emotions [1,20].…”

Adaptive relevance feedback for large-scale image retrieval

“…Movie retrieval could be an interesting application, as raw textual information can be derived from the speech transcript [58] and this textual information could complement greatly the visual content. Songs with lyrics in music databases constitute a similar case [47]. Another useful application could be for the medical databases that associate x-ray images and reports.…”

Learning Timbre Analogies from Unlabelled Data by Multivariate Tree Regression

“…Many musical applications of machine learning can be described as automatic classification tasks, where the class labels may for example indicate genre, key or instrumentation [Orio, 2006]. Less attention has been paid to regressiontype tasks in music processing, yet they too may be facilitated or automated by machine learning.…”