Inter genre similarity modelling for automatic music genre classification
Proceedings of the 9th International Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18–20, 2006, pp. 153–156
Abstract
An abstract is not available in the lab archive. View the paper at its source.
This 2006 conference paper examines whether modelling similarities among music genres can reduce classification confusion. It presents inter-genre similarity, iterative modelling and score-based elimination approaches, evaluated on a nine-genre music database.
Summary
The paper investigates inter-genre similarity modelling (IGS) for automatic music genre classification. The approach models feature frames that were misclassified during training as a separate similarity class, then excludes frames matching that class from genre decisions. The authors also propose iterative IGS and score modelling for IGS elimination (SMIGS).
The evaluation used 566 30-second audio segments across nine genres, with two-fold cross-validation and Gaussian mixture model classifiers. The paper reports average correct-classification rates across decision-window sizes and classifier configurations. These results are from the study’s described database and evaluation setup; they do not establish performance on other datasets or settings.
- Authors: Ulaş Bağcı; Engin Erzin
- Venue: Proceedings of the 9th International Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18–20, 2006, pp. 153–156.
Cite this work
Ulaş Bağcı; Engin Erzin (2006). Inter genre similarity modelling for automatic music genre classification. Proceedings of the 9th International Conference on Digital Audio Effects (DAFx-06), Montreal, Canada, September 18–20, 2006, pp. 153–156 https://www.dafx.de/paper-archive/2006/papers/p_153.pdf
BibTeX