Exploring efficient attention strategies in conformer-based sound event detection

Sound Event Detection (SED) requires models that can accurately localize and classify overlapping audio events within complex acoustic environments. Conformer-based architectures have demonstrated promising performance by leveraging self-attention to capture long-range dependencies. However, this gl...

Descripción completa

Detalles Bibliográficos
Autores: Barahona Quirós, Sara, Álvarez Trejos, Juan Ignacio, Lozano Díez, Alicia, Ramos Castro, Daniel, Torre Toledano, Doroteo
Tipo de recurso: artículo
Fecha de publicación:2026
País:España
Institución:Universidad Autónoma de Madrid
Repositorio:Biblos-e Archivo. Repositorio Institucional de la UAM
Idioma:inglés
OAI Identifier:oai:repositorio.uam.es:10486/754400
Acceso en línea:https://hdl.handle.net/10486/754400
https://dx.doi.org/10.1016/j.csl.2026.101967
Access Level:acceso abierto
Palabra clave:Sound event detection
Efficient conformer
Attention mechanisms
DCASE
Temporal resolution
Informática
id ES_d8cb755dca4fa62dbad3dac12cb41bc9
oai_identifier_str oai:repositorio.uam.es:10486/754400
network_acronym_str ES
network_name_str España
repository_id_str
spelling Exploring efficient attention strategies in conformer-based sound event detectionBarahona Quirós, SaraÁlvarez Trejos, Juan IgnacioLozano Díez, AliciaRamos Castro, DanielTorre Toledano, DoroteoSound event detectionEfficient conformerAttention mechanismsDCASETemporal resolutionInformáticaSound Event Detection (SED) requires models that can accurately localize and classify overlapping audio events within complex acoustic environments. Conformer-based architectures have demonstrated promising performance by leveraging self-attention to capture long-range dependencies. However, this global attention can be accumulated across layers, which can blur local temporal boundaries and reduce detection accuracy, especially for short or closely spaced events. While increasing the input sequence length can help recover temporal detail, the quadratic complexity of Conformers’ self-attention significantly increases computational costs. To address this, we propose integrating the Efficient Conformer architecture, which introduces subsampling along the input sequence length, effectively reducing the temporal dimension within blocks. This design enables processing longer input sequences at finer temporal resolution, enhancing localization accuracy without extending output length. Using the DCASE Challenge 2023 Task 4 benchmark, system performance is evaluated via the threshold-independent Polyphonic Sound Detection Score (PSDS), measuring both localization precision (PSDS1) and class robustness (PSDS2). Experiments on the DESED validation dataset demonstrate that the Efficient Conformer not only improves temporal resolution and long-range dependency modeling, but also outperforms standard Conformer and Convolutional Recurrent Neural Network (CRNN) baselines in PSDS2. Additionally, we explore lightweight attention mechanisms employing squeeze-and-excitation blocks to emulate frequency-axis translation invariance of Frequency Dynamic Convolutions (FDY). Our approach achieves performance comparable to heavier models like FDY+Conformer, while reducing computational cost by over 69%, showing promising results for Conformer-based systems in terms of precision and model efficiencyThis research was supported by FPI PRE2022-104808 funded by FSE, Spain as well as project PID2021-125943OB-I00 funded by MCIN/AEI/10.13039/501100011033/FEDER, UE and project PID2024-160789OB-I00 funded by MICIU/AEI/ 10.13039/501100011033/FEDER, UEElsevierDepartamento de Tecnología Electrónica y de las ComunicacionesEscuela Politécnica SuperiorGobierno de España20262026-03-04research articlehttp://purl.org/coar/resource_type/c_2df8fbb1VoRhttp://purl.org/coar/version/c_970fb48d4fbd8a85info:eu-repo/semantics/articleapplication/pdfhttps://hdl.handle.net/10486/754400https://dx.doi.org/10.1016/j.csl.2026.101967reponame:Biblos-e Archivo. Repositorio Institucional de la UAMinstname:Universidad Autónoma de MadridInglésengopen accesshttp://purl.org/coar/access_right/c_abf2Attribution 4.0 Internationalhttp://creativecommons.org/licenses/by/4.0/info:eu-repo/semantics/openAccessoai:repositorio.uam.es:10486/7544002026-06-23T12:46:27Z
dc.title.none.fl_str_mv Exploring efficient attention strategies in conformer-based sound event detection
title Exploring efficient attention strategies in conformer-based sound event detection
spellingShingle Exploring efficient attention strategies in conformer-based sound event detection
Barahona Quirós, Sara
Sound event detection
Efficient conformer
Attention mechanisms
DCASE
Temporal resolution
Informática
title_short Exploring efficient attention strategies in conformer-based sound event detection
title_full Exploring efficient attention strategies in conformer-based sound event detection
title_fullStr Exploring efficient attention strategies in conformer-based sound event detection
title_full_unstemmed Exploring efficient attention strategies in conformer-based sound event detection
title_sort Exploring efficient attention strategies in conformer-based sound event detection
dc.creator.none.fl_str_mv Barahona Quirós, Sara
Álvarez Trejos, Juan Ignacio
Lozano Díez, Alicia
Ramos Castro, Daniel
Torre Toledano, Doroteo
author Barahona Quirós, Sara
author_facet Barahona Quirós, Sara
Álvarez Trejos, Juan Ignacio
Lozano Díez, Alicia
Ramos Castro, Daniel
Torre Toledano, Doroteo
author_role author
author2 Álvarez Trejos, Juan Ignacio
Lozano Díez, Alicia
Ramos Castro, Daniel
Torre Toledano, Doroteo
author2_role author
author
author
author
dc.contributor.none.fl_str_mv Departamento de Tecnología Electrónica y de las Comunicaciones
Escuela Politécnica Superior
Gobierno de España
dc.subject.none.fl_str_mv Sound event detection
Efficient conformer
Attention mechanisms
DCASE
Temporal resolution
Informática
topic Sound event detection
Efficient conformer
Attention mechanisms
DCASE
Temporal resolution
Informática
description Sound Event Detection (SED) requires models that can accurately localize and classify overlapping audio events within complex acoustic environments. Conformer-based architectures have demonstrated promising performance by leveraging self-attention to capture long-range dependencies. However, this global attention can be accumulated across layers, which can blur local temporal boundaries and reduce detection accuracy, especially for short or closely spaced events. While increasing the input sequence length can help recover temporal detail, the quadratic complexity of Conformers’ self-attention significantly increases computational costs. To address this, we propose integrating the Efficient Conformer architecture, which introduces subsampling along the input sequence length, effectively reducing the temporal dimension within blocks. This design enables processing longer input sequences at finer temporal resolution, enhancing localization accuracy without extending output length. Using the DCASE Challenge 2023 Task 4 benchmark, system performance is evaluated via the threshold-independent Polyphonic Sound Detection Score (PSDS), measuring both localization precision (PSDS1) and class robustness (PSDS2). Experiments on the DESED validation dataset demonstrate that the Efficient Conformer not only improves temporal resolution and long-range dependency modeling, but also outperforms standard Conformer and Convolutional Recurrent Neural Network (CRNN) baselines in PSDS2. Additionally, we explore lightweight attention mechanisms employing squeeze-and-excitation blocks to emulate frequency-axis translation invariance of Frequency Dynamic Convolutions (FDY). Our approach achieves performance comparable to heavier models like FDY+Conformer, while reducing computational cost by over 69%, showing promising results for Conformer-based systems in terms of precision and model efficiency
publishDate 2026
dc.date.none.fl_str_mv 2026
2026-03-04
dc.type.none.fl_str_mv research article
http://purl.org/coar/resource_type/c_2df8fbb1
VoR
http://purl.org/coar/version/c_970fb48d4fbd8a85
dc.type.openaire.fl_str_mv info:eu-repo/semantics/article
format article
dc.identifier.none.fl_str_mv https://hdl.handle.net/10486/754400
https://dx.doi.org/10.1016/j.csl.2026.101967
url https://hdl.handle.net/10486/754400
https://dx.doi.org/10.1016/j.csl.2026.101967
dc.language.none.fl_str_mv Inglés
eng
language_invalid_str_mv Inglés
language eng
dc.rights.none.fl_str_mv open access
http://purl.org/coar/access_right/c_abf2
Attribution 4.0 International
http://creativecommons.org/licenses/by/4.0/
dc.rights.openaire.fl_str_mv info:eu-repo/semantics/openAccess
rights_invalid_str_mv open access
http://purl.org/coar/access_right/c_abf2
Attribution 4.0 International
http://creativecommons.org/licenses/by/4.0/
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
dc.publisher.none.fl_str_mv Elsevier
publisher.none.fl_str_mv Elsevier
dc.source.none.fl_str_mv reponame:Biblos-e Archivo. Repositorio Institucional de la UAM
instname:Universidad Autónoma de Madrid
instname_str Universidad Autónoma de Madrid
reponame_str Biblos-e Archivo. Repositorio Institucional de la UAM
collection Biblos-e Archivo. Repositorio Institucional de la UAM
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869421230025605120
score 15,811543

Similares en LA Referencia