A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features

Bob L. Sturm; Corey Kereliuk; Aggelos Pikrakis

doi:10.1109/CIP.2014.6844511

A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features

Bob L. Sturm, Corey Kereliuk, Aggelos Pikrakis

Institut for Arkitektur og Medieteknologi

Publikation: Bidrag til bog/antologi/rapport/konference proceeding › Konferenceartikel i proceeding › Forskning › peer review

9 Citationer (Scopus)

Abstract

Systems built using deep learning neural networks trained on low-level spectral periodicity features (DeSPerF) reproduced the most “ground truth” of the systems submitted to the MIREX 2013 task, “Audio Latin Genre Classification.” To answer why this was the case, we take a closer look at the behavior of a DeSPerF system we create and evaluate using the benchmark dataset BALLROOM. We find through time stretching that this DeSPerF system appears to obtain a high figure of merit on the task of music genre recognition because of a confounding of tempo with “ground truth” in BALLROOM. This observation motivates several predictions.

Originalsprog	Engelsk
Titel	Proceedings of 4th International Workshop on Cognitive Information Processing (CIP 2014)
Redaktører	Lars Kai Hansen, Søren Holdt Jensen, Jan Larsen
Antal sider	6
Vol/bind	1
Forlag	IEEE (Institute of Electrical and Electronics Engineers)
Publikationsdato	2014
Sider	1-6
ISBN (Trykt)	978-1-4799-3696-0
DOI	https://doi.org/10.1109/CIP.2014.6844511
Status	Udgivet - 2014
Begivenhed	International Workshop on Cognitive Information Processing - Bella Sky Hotel, Copenhagen, Danmark Varighed: 26 maj 2014 → 28 maj 2014

Konference

Konference	International Workshop on Cognitive Information Processing
Lokation	Bella Sky Hotel
Land/Område	Danmark
By	Copenhagen
Periode	26/05/2014 → 28/05/2014

Adgang til dokumentet

10.1109/CIP.2014.6844511

AUB Link

Søg efter materialet i Aalborg Universitetsbiblioteks søgemaskine

CoSound
Christensen, M. G. (Projektdeltager), Tan, Z.-H. (Projektdeltager), Jensen, S. H. (Projektdeltager) & Sturm, B. L. (Projektdeltager)
01/01/2012 → 31/12/2015
Projekter: Projekt › Forskning

Citationsformater

Sturm, B. L., Kereliuk, C., & Pikrakis, A. (2014). A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features. I L. K. Hansen, S. Holdt Jensen, & J. Larsen (red.), Proceedings of 4th International Workshop on Cognitive Information Processing (CIP 2014) (Bind 1, s. 1-6). IEEE (Institute of Electrical and Electronics Engineers). https://doi.org/10.1109/CIP.2014.6844511

Sturm, Bob L. ; Kereliuk, Corey ; Pikrakis, Aggelos. / A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features. Proceedings of 4th International Workshop on Cognitive Information Processing (CIP 2014). red. / Lars Kai Hansen ; Søren Holdt Jensen ; Jan Larsen. Bind 1 IEEE (Institute of Electrical and Electronics Engineers), 2014. s. 1-6

@inproceedings{f093c813f3c64c4fb9a4515ab720ff45,

title = "A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features",

abstract = "Systems built using deep learning neural networks trained on low-level spectral periodicity features (DeSPerF) reproduced the most “ground truth” of the systems submitted to the MIREX 2013 task, “Audio Latin Genre Classification.” To answer why this was the case, we take a closer look at the behavior of a DeSPerF system we create and evaluate using the benchmark dataset BALLROOM. We find through time stretching that this DeSPerF system appears to obtain a high figure of merit on the task of music genre recognition because of a confounding of tempo with “ground truth” in BALLROOM. This observation motivates several predictions.",

author = "Sturm, {Bob L.} and Corey Kereliuk and Aggelos Pikrakis",

year = "2014",

doi = "10.1109/CIP.2014.6844511",

language = "English",

isbn = "978-1-4799-3696-0",

volume = "1",

pages = "1--6",

editor = "Hansen, {Lars Kai} and {Holdt Jensen}, {S{\o}ren } and Jan Larsen",

booktitle = "Proceedings of 4th International Workshop on Cognitive Information Processing (CIP 2014)",

publisher = "IEEE (Institute of Electrical and Electronics Engineers)",

address = "United States",

note = "International Workshop on Cognitive Information Processing ; Conference date: 26-05-2014 Through 28-05-2014",

}

Sturm, BL, Kereliuk, C & Pikrakis, A 2014, A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features. i LK Hansen, S Holdt Jensen & J Larsen (red), Proceedings of 4th International Workshop on Cognitive Information Processing (CIP 2014). bind 1, IEEE (Institute of Electrical and Electronics Engineers), s. 1-6, International Workshop on Cognitive Information Processing, Copenhagen, Danmark, 26/05/2014. https://doi.org/10.1109/CIP.2014.6844511

A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features. / Sturm, Bob L.; Kereliuk, Corey; Pikrakis, Aggelos.
Proceedings of 4th International Workshop on Cognitive Information Processing (CIP 2014). red. / Lars Kai Hansen; Søren Holdt Jensen; Jan Larsen. Bind 1 IEEE (Institute of Electrical and Electronics Engineers), 2014. s. 1-6.

Publikation: Bidrag til bog/antologi/rapport/konference proceeding › Konferenceartikel i proceeding › Forskning › peer review

TY - GEN

T1 - A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features

AU - Sturm, Bob L.

AU - Kereliuk, Corey

AU - Pikrakis, Aggelos

PY - 2014

Y1 - 2014

N2 - Systems built using deep learning neural networks trained on low-level spectral periodicity features (DeSPerF) reproduced the most “ground truth” of the systems submitted to the MIREX 2013 task, “Audio Latin Genre Classification.” To answer why this was the case, we take a closer look at the behavior of a DeSPerF system we create and evaluate using the benchmark dataset BALLROOM. We find through time stretching that this DeSPerF system appears to obtain a high figure of merit on the task of music genre recognition because of a confounding of tempo with “ground truth” in BALLROOM. This observation motivates several predictions.

AB - Systems built using deep learning neural networks trained on low-level spectral periodicity features (DeSPerF) reproduced the most “ground truth” of the systems submitted to the MIREX 2013 task, “Audio Latin Genre Classification.” To answer why this was the case, we take a closer look at the behavior of a DeSPerF system we create and evaluate using the benchmark dataset BALLROOM. We find through time stretching that this DeSPerF system appears to obtain a high figure of merit on the task of music genre recognition because of a confounding of tempo with “ground truth” in BALLROOM. This observation motivates several predictions.

U2 - 10.1109/CIP.2014.6844511

DO - 10.1109/CIP.2014.6844511

M3 - Article in proceeding

SN - 978-1-4799-3696-0

VL - 1

SP - 1

EP - 6

BT - Proceedings of 4th International Workshop on Cognitive Information Processing (CIP 2014)

A2 - Hansen, Lars Kai

A2 - Holdt Jensen, Søren

A2 - Larsen, Jan

PB - IEEE (Institute of Electrical and Electronics Engineers)

T2 - International Workshop on Cognitive Information Processing

Y2 - 26 May 2014 through 28 May 2014

ER -

Sturm BL, Kereliuk C, Pikrakis A. A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features. I Hansen LK, Holdt Jensen S, Larsen J, red., Proceedings of 4th International Workshop on Cognitive Information Processing (CIP 2014). Bind 1. IEEE (Institute of Electrical and Electronics Engineers). 2014. s. 1-6 doi: 10.1109/CIP.2014.6844511

A Closer Look at Deep Learning Neural Networks with Low-level Spectral Periodicity Features

Abstract

Konference

Adgang til dokumentet

AUB Link

Fingeraftryk

Projekter

CoSound

Citationsformater