Adversarially Training for Audio Classifiers

TL;DR

This paper explores how adversarial training affects the robustness of deep neural networks for audio classification, showing improved resistance to attacks but with some accuracy trade-offs across two sound datasets.

Contribution

It demonstrates the effectiveness of adversarial training in increasing robustness of audio classifiers against various attacks, highlighting the trade-offs involved.

Findings

Adversarial training reduces recognition accuracy.

Adversarially trained models can have over 90% fooling rate.

Higher perturbations are needed to fool adversarially trained models.

Abstract

In this paper, we investigate the potential effect of the adversarially training on the robustness of six advanced deep neural networks against a variety of targeted and non-targeted adversarial attacks. We firstly show that, the ResNet-56 model trained on the 2D representation of the discrete wavelet transform appended with the tonnetz chromagram outperforms other models in terms of recognition accuracy. Then we demonstrate the positive impact of adversarially training on this model as well as other deep architectures against six types of attack algorithms (white and black-box) with the cost of the reduced recognition accuracy and limited adversarial perturbation. We run our experiments on two benchmarking environmental sound datasets and show that without any imposed limitations on the budget allocations for the adversary, the fooling rate of the adversarially trained models can exceed 90\%. In other words, adversarial attacks exist in any scales, but they might require higher adversarial perturbations compared to non-adversarially trained models.