ComSense-CNN: acoustic event classification via 1D convolutional neural network with compressed sensing

Citation

Tan, Pooi Shiang and Lim, Kian Ming and Tan, Cheah Heng and Lee, Chin Poo and Kwek, Lee Chung (2022) ComSense-CNN: acoustic event classification via 1D convolutional neural network with compressed sensing. Signal, Image and Video Processing. ISSN 1863-1703

[img] Text
8.pdf - Published Version
Restricted to Repository staff only

Download (883kB)

Abstract

Sound plays an important role in human daily life as humans use sound to communicate with each other and to understand the events occurring in the surroundings. This has prompted the researchers to further study on how to automatically identify the event that is happening by analyzing the acoustic signal. This paper presents a deep learning model enhanced by compressed sensing techniques for acoustic event classification. The compressed sensing first transforms the input acoustic signal into a reconstructed signal to reduce the noise in the input acoustic signal. The reconstructed signals are then fed into a 1-dimensional convolutional neural network (1D-CNN) to train a deep learning model for the acoustic event classification. In addition, the dropout regularization is leveraged in the 1D-CNN to mitigate the overfitting problems. The proposed compressed sensing with 1D-CNN was evaluated on three benchmark datasets, namely Soundscapes1, Soundscapes2, and UrbanSound8K, and achieved F1-scores of 80.5%, 81.1%, and 69.2%, respectively.

Item Type: Article
Uncontrolled Keywords: Convolutional neural network, Neural network, deep learning
Subjects: Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions: Faculty of Engineering and Technology (FET)
Faculty of Information Science and Technology (FIST)
Depositing User: Ms Nurul Iqtiani Ahmad
Date Deposited: 01 Aug 2022 00:48
Last Modified: 01 Aug 2022 00:48
URII: http://shdl.mmu.edu.my/id/eprint/10253

Downloads

Downloads per month over past year

View ItemEdit (login required)