Use this URL to cite or link to this record in EThOS:
Title: Content based retrieval and classification of music using polyphonic timbre similarity
Author: De Leon, Franz
ISNI:       0000 0004 5347 1279
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2014
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Digital technology and the Internet have changed the music industry's landscape. Music has become more accessible allowing consumers to store and share thousands of items in their computer's hard disk, portable media player, mobile phoone and other devices. Recent developments allow consumers to store digital music on the Internet through cloud storage. Given the large music collections available, there is a need for new applications for browsing, organising, discovering and generating playlists for users. In previous years, searching for music has been similar to a textual information search. However, this limits music discovery as it usually requires specific information that may be unknown to the user. This thesis investigates three of the core components of content-based music retrieval: audio features, similarity functions and indexing methods. In the content-based paradigm, audio files are analyzed using their waveform and are represented by high-dimensional features. This study focuses on polyphonic timbre similarity. Polyphonic timbre is the characteristic that allows listeners to differentiate between two music signals or complex instrumental textures with the same perceived pitch and loudness. The different attributes of timbre are examined and suitable features that can be used for music retrieval using timbre similarity are investigated. Evaluations are performed to compare the performance of these features. To improve the overall performance and reduce the undesirable effects of operating in high-dimensionality space, methods on how feature spaces can be combined are also explored. A full linear scan of the feature space is impractical for large music collections. Hence, the filter-and-refine method is adopted to expedite the retrieval process. The objective is to filter a dataset by quickly returning a set of candidate songs then refining the results using an exact similarity measure. Some novel modifications of the filtering step are made to ensure that the level of performance is maintained. The application of our timbre similarity systems are extended to automatic audio classification. In the paradigm, anunlabeled track is tagged with the label of the nearest track. Finally, the performance of our similarity estimator and audio classifier are validated in the annual Music Information Retrieval Evaluation eXchange (MIREX). The MIREX results show that our techniques are state-of-the-art methods.
Supervisor: Martinez, Kirk Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: M Music ; QA75 Electronic computers. Computer science