Accumulative feature extracting network for sound event detection

Sangwon Park; Sangwook Park

doi:10.7776/ASK.2025.44.5.489

All Issue

2025 Vol.44, Issue 5 Preview Page Next Page

Research Article

Accumulative feature extracting network for sound event detection 음향 이벤트 검출을 위한 누적 특징 추출 네트워크

30 September 2025. pp. 489-495

PDF XML

Abstract

Sound event detection is a technology that detects the type, onset, and offset of sound events in audio signals and it is used in various fields such as monitoring systems and autonomous vehicles. Through the international competition (Detection and Classification of Acoustic Scenes and Events, DCASE) on acoustic signal analysis , various methods have been introduced to improve the performance of sound event detection. In this paper, we propose AccNet to solve the loss of spectro-temporal information in low layers of conventional acoustic model. In experiments performed on the DCASE 2023 Task 4 testbed, while the proposed model is comparable to the DCASE 2023 baseline in model complexity, it achieved 44.76 ± 0.51 % in event based f1 score, the best performance compared to the other models such as CRNN, multi-resolutional convolution based model, and residual path based model. Also the proposed model demonstrates improved f1 scores for target sound event except Blender and Electric shaver, compared to the residual path based model.

Keywords

Sound event detection

Detection and Classification of Acoustic Scenes and Events (DCASE) 2023 Task4

Dense Net

Network architecture

Temporal embedding

음향 이벤트 검출은 오디오 신호에서 음향의 종류와 발생 지점과 끝점을 검출하는 기술로 모니터링 시스템, 자율주행 자동차 등 다양한 분야에 쓰이고 있다. 음향 이벤트 검출은 음향 신호 분석에 관한 국제 경연대회(Detection and Classification of Acoustic Scenes and Events, DCASE)를 통해 음향 이벤트 검출 성능을 향상시키기 위한 다양한 방법들이 소개되고 있다. 본 논문은 기존 음향 분석 모델의 하위 계층에서 시간-주파수 정보가 손실되는 문제를 완화하기 위해 누적 특성 추출 신경망(AccNet)을 제안한다. 제안하는 모델은 DCASE 2023 task4 테스트 베드를 활용한 실험에서, DCASE 2023 Baseline과 동일한 파라미터 수를 유지하면서 F1 점수에서 44.76 ± 0.51[%]를 기록하였고, 비교대상으로 고려된 CRNN, 다중해상도 합성곱 모델, 잔차 경로 기반 모델들에 비해 가장 우수한 성능을 보여준다. 또한, 제안하는 모델은 잔차 경로 기반 모델에 비해 Blender와 Electric shaver를 제외한 관심 음향에서 향상된 F1 점수를 보여준다.

키워드

음향 이벤트 검출

Detection and Classification of Acoustic Scenes and Events (DCASE) 2023 Task4

밀집 신경망

네트워크 구조

시계열 임베딩

References

G. Ciaburro and G. Iannace, “Improving smart cities safety using sound events detection based on deep neural network algorithms,” Informatics, 7, 23 (2020).

10.3390/informatics7030023

A. H. Yuh and S. J. Kang, “Real-time sound event classification for human activity of daily living using deep neural network,” Proc. IEEE iThings, GreenCom, CPSCom, SmartData, Cybermatics, 83-88 (2021).

10.1109/iThings-GreenCom-CPSCom-SmartData-Cybermatics53846.2021.00027

H. G. Kim and G. Y. Kim, “Deep neural network- based indoor emergency awareness using contextual information from sound, human activity, and indoor position on mobile device,” IEEE Trans. Consum. Electron. 66, 271-278 (2020).

10.1109/TCE.2020.3015197

P. Giannakopoulos, A. Pikrakis, and Y. Cotronis, “Improving post-processing of audio event detectors using reinforcement learning,” IEEE Access, 10, 84398–84404 (2022).

10.1109/ACCESS.2022.3197907

D. de Benito-Gorrón, D. Ramos, and D. T. Toledano, “A multi-resolution CRNN-based approach for semi- supervised sound event detection in DCASE 2020 challenge,” IEEE Access, 9, 89029-89042 (2021).

10.1109/ACCESS.2021.3088949

J. Nam and S. W. Park, “Boosting principal frequency based data augmentation for sound event detection” (in Korean), J. Korean Inst. Electron. Eng. 77-83 (2024).

10.5573/ieie.2024.61.7.77

N. K. Kim and H. K. Kim, “Self-training with noisy student model and semi-supervised loss function for DCASE 2021 challenge task 4,” DCASE, Tech. Rep., 2021.

L. Lin, X. Wang, H. Liu, and Y. L. Qian, “Guided learning convolution system for dcase 2019 task 4,” arXiv preprint arXiv:1909.06178 (2019).

10.33682/53ed-z889

C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking the Inception architecture for computer vision,” Proc. IEEE CVPR, 2818-2826 (2016).

10.1109/CVPR.2016.308

W. Lim, S. Suh, and Y. Jeong, “Weakly labeled semi-supervised sound event detection using CRNN with inception module,” Proc. DCASE, 74-77 (2018).

K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” Proc. IEEE CVPR, 770-778 (2016).

10.1109/CVPR.2016.90

G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” Proc. IEEE CVPR, 4700-4708 (2017).

10.1109/CVPR.2017.243

G. Huang, S. Liu, L. Van Der Maaten, and K. Q. Weinberger, “CondenseNet: An efficient DenseNet using learned group convolutions,” Proc. IEEE CVPR, 2752-2761 (2018).

10.1109/CVPR.2018.00291

N. Turpault, R. Serizel, A. Shah, and J. Salamon, “Sound event detection in domestic environments with weakly labeled data and soundscape synthesis,” Proc. DCASE, Workshop, 253-257 (2019).

10.33682/006b-jx26

Ç. Bilen, G. Ferroni, F. Tuveri, J. Azcarreta, and S. Krstulović, “A framework for the robust evaluation of sound event detection,” Proc. IEEE ICASSP, 61-65 (2020).

10.1109/ICASSP40776.2020.9052995

DCASE Community, https://dcase.community/challenge2023/task-sound-event-detection-with-weak-labels-and-synthetic-soundscapes-results#task-description, (Last viewed April 15, 2025).

H. Nam, S. H. Kim, B. Y. Ko, and Y. H. Park, “Frequency dynamic convolution: Frequency-adaptive pattern recognition for sound event detection,” arXiv preprint arXiv:2203.15296 (2022).

10.21437/Interspeech.2022-10127

T. Song and W. Zhang, “Frequency-aware convolution for sound event detection,” Proc. ICMM, 415- 426 (2025).

10.1007/978-981-96-2054-8_31

Information

Publisher :The Acoustical Society of Korea
Publisher(Ko) :한국음향학회
Journal Title :The Journal of the Acoustical Society of Korea
Journal Title(Ko) :한국음향학회지
Volume : 44
No :5
Pages :489-495
Received Date : 2025-05-26
Revised Date : 2025-07-07
Accepted Date : 2025-07-20
DOI :https://doi.org/10.7776/ASK.2025.44.5.489

The Journal of the Acoustical Society of KoreaISSN:1225-4428(Print) 2287-3775(Online)한국음향학회

All Issue