Methodological Considerations Concerning Manual Annotation Of Musical Audio In Function Of Algorithm Development

In research on musical audio-mining, annotated music databases are needed which allow the development of computational tools that extract from the musical audiostream the kind of high-level content that users can deal with in Music Information Retrieval (MIR) contexts. The notion of musical content, and therefore the notion of annotation, is ill-defined, however, both in the syntactic and semantic sense. As a consequence, annotation has been approached from a variety of perspectives (but mainly linguistic-symbolic oriented), and a general methodology is lacking. This paper is a step towards the definition of a general framework for manual annotation of musical audio in function of a computational approach to musical audio-mining that is based on algorithms that learn from annotated data.

• Type: Conference
• Year: 2004
• Where: ISMIR
• Authors: Micheline Lesaffre, Marc Leman, Bernard De Baets, Jean-Pierre Martens
• Keywords: ISMIR 2004, Music Information Retrieval, Musical Audio-mining, Musical Audiostream