English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

This paper investigates the impact of different standard environmental soundrepresentations (spectrograms) on the recognition performance and adversarialattack robustness of a victim residual convolutional neural network, namelyResNet-18. Our main motivation for focusing on such a front-end classifierrather than other complex architectures is balancing recognition accuracy andthe total number of training parameters. Herein, we measure the impact ofdifferent settings required for generating more informative Mel-frequencycepstral coefficient (MFCC), short-time Fourier transform (STFT), and discretewavelet transform (DWT) representations on our front-end model. Thismeasurement involves comparing the classification performance over theadversarial robustness. We demonstrate an inverse relationship betweenrecognition accuracy and model robustness against six benchmarking attackalgorithms on the balance of average budgets allocated by the adversary and theattack cost. Moreover, our experimental results have shown that while theResNet-18 model trained on DWT spectrograms achieves a high recognitionaccuracy, attacking this model is relatively more costly for the adversary thanother 2D representations. We also report some results on differentconvolutional neural network architectures such as ResNet-34, ResNet-56,AlexNet, and GoogLeNet, SB-CNN, and LSTM-based.

From Environmental Sound Representation to Robustness of 2D CNN Models Against Adversarial Attacks