?El Caballo Viejo? Latin Genre Recognition with Deep Learning and Spectral Periodicity

Bob L. Sturm, Corey Kereliuk, Jan Larsen

AbstractThe winning" system in the 2013 MIREX Latin Genre Classification Train-test Task was a deep neural network trained with simple periodicity features. The explanation for its winning performance has yet to be fully explained. In our previous work, we built similar systems using the BALLROOM music dataset, and found its performance to be greatly affected by minor time dilation of the input, in effect changing music tempo. In the MIREX task, however, systems are trained and tested using the Latin Music Dataset (LMD), which is 4.5 times larger than BALLROOM and which does not seem to show a strong relation-
ship between tempo and label. In this paper, we reproduce the "winning" deep learning system using LMD, and measure the effects of time dilation on its performance. We find that, just as for BALLROOM, tempo changes of at most 6% greatly hurt or benefit its performance. Interpreted with the low-level nature of the input features, this supports the conclusion that the system is exploiting some low-level absolute time characteristics to reproduce the ground truth in LMD.
TypeConference paper [With referee]
ConferenceFifth Biennial International Conference on Mathematics and Computation in Music (MCM2015)
Year2015    Month June
PublisherSpringer
SeriesLecture Notes in Computer Science
Electronic version(s)[pdf]
Publication linkhttp://mcm2015.qmul.ac.uk/
BibTeX data [bibtex]
IMM Group(s)Intelligent Signal Processing


Back  ::  IMM Publications