Towards computational morphological description of sound

TitleTowards computational morphological description of sound
Publication TypeMaster Thesis
Year of Publication2004
AuthorsRicard, J.
preprint/postprint documentfiles/publications/DEA-Ricard.pdf
AbstractResearch on audio content description deals with limited types of sounds. Most of the work done in this area is applied to automatic transcription of traditional western music, i.e. the conversion of audio into the traditional musical notation pitch/duration/loudness/source or the recognition of the origin of specific sounds (speech, music, applause...) for indexing or retrieval purpose. In that context, electronic sounds, noises or sounds that have no identifiable origin, which are used a lot in contemporary music and sound post production for video or cinema, can hardly be handled. In this document, we propose an alternative representation, inspired by Pierre Schaeffer's work on sound objects, based on a limited number of perceptual criteria that can be applied to any type of sound. More specifically, we describe our first attempt to automatically characterize some of these criteria, called morphological criteria, as well as an evaluation of the usability of the resulting representation in the context of sound retrieval. Conclusions drawn from these experiments to improve and complete the system, as well as a the description of potential applications, are presented as future work to be done in the final thesis.
intranet