Beat Tracking: Is 44.1 kHz Really Needed?
The result's identifiers
Result code in IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26220%2F23%3APU148273" target="_blank" >RIV/00216305:26220/23:PU148273 - isvavai.cz</a>
Result on the web
<a href="https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2023_sbornik_2_v2.pdf" target="_blank" >https://www.eeict.cz/eeict_download/archiv/sborniky/EEICT_2023_sbornik_2_v2.pdf</a>
DOI - Digital Object Identifier
<a href="http://dx.doi.org/10.13164/eeict.2023.227" target="_blank" >10.13164/eeict.2023.227</a>
Alternative languages
Result language
angličtina
Original language name
Beat Tracking: Is 44.1 kHz Really Needed?
Original language description
Beat tracking is essential in music information retrieval, with applications ranging from music analysis and automatic playlist generation to beat-synchronized effects. In recent years, deep learning methods, usually inspired by well-known architectures, outperformed other beat tracking algorithms. The current state-of-the-art offline beat tracking systems utilize temporal convolutional and recurrent networks. Most systems use an input sampling rate of 44.1 kHz. In this paper, we retrain multiple versions of state-of-the-art temporal convolutional networks with different input sampling rates while keeping the time resolution by changing the frame size parameter. Furthermore, we evaluate all models using standard metrics. As the main contribution, we show that decreasing the input audio recording sampling frequency up to 5 kHz preserves most of the accuracy, and in some cases, even slightly outperforms the standard approach.
Czech name
—
Czech description
—
Classification
Type
D - Article in proceedings
CEP classification
—
OECD FORD branch
20203 - Telecommunications
Result continuities
Project
—
Continuities
S - Specificky vyzkum na vysokych skolach
Others
Publication year
2023
Confidentiality
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů
Data specific for result type
Article name in the collection
Proceedings II of the 29th Conference STUDENT EEICT 2023 Selected Papers
ISBN
978-80-214-6154-3
ISSN
—
e-ISSN
—
Number of pages
5
Pages from-to
227-231
Publisher name
Brno University of Technology, Faculty of Elektronic Engineering and Communication
Place of publication
Brno
Event location
Brno
Event date
Apr 25, 2023
Type of event by nationality
WRD - Celosvětová akce
UT code for WoS article
—