Lightweight machine learning models for soundscape source classification in architectural and environmental acoustics: A comparative analysis

Wongeun Oh

doi:10.7776/ASK.2025.44.6.720

All Issue

2025 Vol.44, Issue 6 Preview Page

Research Article

Lightweight machine learning models for soundscape source classification in architectural and environmental acoustics: A comparative analysis 건축·환경음향 응용을 위한 사운드스케이프 음원 분류경량 머신러닝 모델 비교 분석

30 November 2025. pp. 720-728

PDF XML

Abstract

This paper compares and analyzes the performance of lightweight machine learning models for soundscape classification in edge computing environments, with applications in architectural and environmental acoustics. Three CNN-based models and three tree-based models were employed, and the ESC-50 dataset was reorganized into five soundscape classification taxonomies for evaluation. The experimental results showed that tree-based models using YAMNet embeddings achieved higher accuracy than CNN models in most classification schemes. In addition, theoretical estimates of computational efficiency on the Raspberry Pi 5 indicated that this combination achieved the fastest processing times. Therefore, the combination of YAMNet embeddings and tree models is expected to be an effective approach for real-time soundscape source classification on edge devices.

Keywords

Soundscape source classification

Edge computing

Lightweight CNNs

YAMNet embedding

ESC-50

본 논문에서는 건축음향 및 환경음향 분야에 활용 목적으로 엣지 컴퓨팅 환경에서 사운드스케이프 음원 분류 체계 인식을 위한 경량 머신러닝 모델의 성능을 비교 분석하였다. 실험에는 Convolutional Neural Network(CNN) 계열 모델 3종과 트리 기반 모델 3종을 사용하였으며, ESC-50 데이터셋을 다섯 가지 사운드스케이프 음원 분류 체계로 재구성하여 인식 성능을 평가하였다. 실험 결과, YAMNet 임베딩을 사용한 트리 모델이 대부분의 분류 체계에서 CNN 모델보다 높은 정확도를 보였다. 또한 라즈베리파이 5에서의 처리 시간을 이론적으로 계산한 연산 효율성 측면에서도 이 조합이 가장 빠른 처리 속도를 보였다. 따라서 YAMNet 임베딩과 트리 모델의 조합은 엣지 디바이스에서 실시간 사운드스케이프 음원 분류에 효과적인 방안이 될 수 있을 것으로 기대된다.

키워드

사운드스케이프 음원 분류

엣지 컴퓨팅

경량 CNN

YAMNet 임베딩

ESC-50

References

R. M. Schafer, The Soundscape: Our Sonic Environment and the Tuning of the World (Destiny Books, Rochester,1994), pp. 3-8.

ISO 12913-1:2014, Acoustics — Soundscape — Part 1: Definition and conceptual framework, International Organization for Standardization, 2014.

J. Kang, F. Aletta, T. T. Gjestland, L. A. Brown, D. Botteldooren, B. Schulte-Fortkamp, P. Lercher, I. van Kamp, K. Genuit, A. Fiebig, J. L. B. Coelho, L. Maffei, and L. Lavia, “Ten questions on the soundscapes of the built environment,” Build. Environ. 108, 284-294 (2016).

10.1016/j.buildenv.2016.08.011

J. Kang, “From dBA to soundscape indices: Managing our sound environment,” Front. Eng, 4, 184-192, (2017).

10.15302/J-FEM-2017026

H. Bellafkir, M. Vogelbacher, and B. Freisleben, “Urban sound classification on resource-constrained edge devices,” Proc. Int. Conf. Service-Oriented Computing, 237-250 (2024).

10.1007/978-981-96-7238-7_19

H.-I. Liu, D. Wang, C. Xu, Z. Wang, Y. Zhang, and L. Chen, “Lightweight deep learning for resource-constrained environments: A survey,” ACM Comput. Surv. 56, 1-42 (2024).

10.1145/3657282

K. J. Piczak, “ESC: Dataset for environmental sound classification,” Proc. ACM Multimedia, 1015-1018 (2015).

10.1145/2733373.2806390

A. Howard, M. Sandler, G. Chu, L.-C. Chen, B. Chen, M. Tan, W. Wang, Y. Zhu, R. Pang, V. Vasudevan, Q. V. Le, and H. Adam, “Searching for MobileNetV3,” Proc. IEEE/CVF Int. Conf. Computer Vision, 1314-1324 (2019).

10.1109/ICCV.2019.00140

M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “MobileNetV2: Inverted residuals and linear bottlenecks,” Proc. IEEE Conf. Computer Vision and Pattern Recognition, 4510-4520 (2018).

10.1109/CVPR.2018.00474

M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” Proc. Int. Conf. Machine Learning, 6105-6114 (2019).

G. Ke, Q. Meng, T. Finley, T. Wang, W. Chen, W. Ma, Q. Ye, and T.-Y. Liu, “LightGBM: A highly efficient gradient boosting decision tree,” Proc. NeurIPS, 3146-3154 (2017).

T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” Proc. 22nd ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining, 785-794 (2016).

10.1145/2939672.2939785

L. Breiman, “Random Forests,” Mach. Learn. 45, 5-32 (2001).

10.1023/A:1010933404324

Transfer Learning for Audio Data with YAMNet, https://blog.tensorflow.org/2021/03/transfer-learning-for-audio-data-with-yamnet.html, (Last viewed November 1, 2025).

ISO/TS 12913-2:2018, Acoustics - Soundscape - Part 2: Data collection and reporting, International Organization for Standardization, 2018.

B. L. Krause, The Great Animal Orchestra: Finding The Origins of Music in the World’s Wild Places (Little, Brown and Company, New York, 2012), pp. 52-53.

B. C. Pijanowski, L. J. Villanueva-Rivera, S. L. Dumyahn, A. Farina, B. L. Krause, B. M. Napoletano, S. H. Gage, and N. Pieretti, “Soundscape ecology: the science of sound in the landscape,” BioScience, 61, 203-216 (2011).

10.1525/bio.2011.61.3.6

T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, “Optuna: A next-generation hyperparameter optimization framework,” Proc. 25th ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining, 2623-2631 (2019).

10.1145/3292500.3330701

The GFLOPS/W of the Various Machines in the VMW Research Group, https://web.eece.maine.edu/~vweaver/group/green_machines.html, (Last viewed September 17, 2025).

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 :6
Pages :720-728
Received Date : 2025-09-17
Accepted Date : 2025-11-05
DOI :https://doi.org/10.7776/ASK.2025.44.6.720

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

All Issue