Shortest path techniques for annotation and retrieval of environmental sounds

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Scopus citations

Abstract

Many techniques for text-based retrieval and automatic annotation of music and sound effects rely on learning with explicit generalization, training individual classifiers for each tag. Non-parametric approaches, where queries are individually compared to training instances, can provide added flexibility, both in terms of robustness to shifts in database content and support for foreign queries, such as concepts not yet included in the database. In this paper, we build upon prior work in designing an ontological framework for annotation and retrieval of environmental sounds, where shortest paths are used to navigate a network containing edges that represent content-based similarity, semantic similarity, and user tagging data. We evaluate novel techniques for ordering query results using weights of both shortest paths and minimum cost paths of specified lengths, pruning outbound edges by nodes' K nearest neighbors, and adjusting edge weights depending on type (acoustic, semantic, or user tagging). We evaluate these methods both through traditional cross-validation and through simulation of live systems containing a complete collection of sounds and tags but incomplete tagging data.

Original languageEnglish (US)
Title of host publicationProceedings of the 13th International Society for Music Information Retrieval Conference, ISMIR 2012
Pages541-546
Number of pages6
StatePublished - Dec 1 2012
Event13th International Society for Music Information Retrieval Conference, ISMIR 2012 - Porto, Portugal
Duration: Oct 8 2012Oct 12 2012

Publication series

NameProceedings of the 13th International Society for Music Information Retrieval Conference, ISMIR 2012

Other

Other13th International Society for Music Information Retrieval Conference, ISMIR 2012
CountryPortugal
CityPorto
Period10/8/1210/12/12

ASJC Scopus subject areas

  • Music
  • Information Systems

Fingerprint Dive into the research topics of 'Shortest path techniques for annotation and retrieval of environmental sounds'. Together they form a unique fingerprint.

  • Cite this

    Mechtley, B., Cook, P., & Spanias, A. (2012). Shortest path techniques for annotation and retrieval of environmental sounds. In Proceedings of the 13th International Society for Music Information Retrieval Conference, ISMIR 2012 (pp. 541-546). (Proceedings of the 13th International Society for Music Information Retrieval Conference, ISMIR 2012).