Saved in:
Bibliographic Details
Main Authors: Wang, Qing, Jiang, Ya, Chen, Hang, Siniscalchi, Sabato Marco, Du, Jun, Gao, Jianqing
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.12334
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908562581618688
author Wang, Qing
Jiang, Ya
Chen, Hang
Siniscalchi, Sabato Marco
Du, Jun
Gao, Jianqing
author_facet Wang, Qing
Jiang, Ya
Chen, Hang
Siniscalchi, Sabato Marco
Du, Jun
Gao, Jianqing
contents This work presents HDA-SELD, a unified framework that combines hierarchical cross-modal distillation (HCMD) and multi-level data augmentation to address low-resource audio-visual (AV) sound event localization and detection (SELD). An audio-only SELD model acts as the teacher, transferring knowledge to an AV student model through both output responses and intermediate feature representations. To enhance learning, data augmentation is applied by mixing features randomly selected from multiple network layers and associated loss functions tailored to the SELD task. Extensive experiments on the DCASE 2023 and 2024 Challenge SELD datasets show that the proposed method significantly improves AV SELD performance, yielding relative gains of 21%-38% in the overall metric over the baselines. Notably, our proposed HDA-SELD achieves results comparable to or better than teacher models trained on much larger datasets, surpassing state-of-the-art methods on both DCASE 2023 and 2024 Challenge SELD tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2508_12334
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle HDA-SELD: Hierarchical Cross-Modal Distillation with Multi-Level Data Augmentation for Low-Resource Audio-Visual Sound Event Localization and Detection
Wang, Qing
Jiang, Ya
Chen, Hang
Siniscalchi, Sabato Marco
Du, Jun
Gao, Jianqing
Sound
Multimedia
This work presents HDA-SELD, a unified framework that combines hierarchical cross-modal distillation (HCMD) and multi-level data augmentation to address low-resource audio-visual (AV) sound event localization and detection (SELD). An audio-only SELD model acts as the teacher, transferring knowledge to an AV student model through both output responses and intermediate feature representations. To enhance learning, data augmentation is applied by mixing features randomly selected from multiple network layers and associated loss functions tailored to the SELD task. Extensive experiments on the DCASE 2023 and 2024 Challenge SELD datasets show that the proposed method significantly improves AV SELD performance, yielding relative gains of 21%-38% in the overall metric over the baselines. Notably, our proposed HDA-SELD achieves results comparable to or better than teacher models trained on much larger datasets, surpassing state-of-the-art methods on both DCASE 2023 and 2024 Challenge SELD tasks.
title HDA-SELD: Hierarchical Cross-Modal Distillation with Multi-Level Data Augmentation for Low-Resource Audio-Visual Sound Event Localization and Detection
topic Sound
Multimedia
url https://arxiv.org/abs/2508.12334