RSD-GAN: Regularized Sobolev Defense GAN Against Speech-to-Text Adversarial Attacks

TL;DR

This paper presents RSD-GAN, a novel Sobolev-based GAN with regularization that enhances defense against diverse adversarial attacks on speech-to-text systems, demonstrating superior robustness in experiments.

Contribution

Introduction of RSD-GAN, a regularized Sobolev GAN, as an effective defense mechanism for speech-to-text adversarial attacks, with a new regularizer controlling the discriminator's functionality.

Findings

RSD-GAN significantly improves robustness against targeted and non-targeted attacks.

Experimental results on DeepSpeech, Kaldi, and Lingvo show superior defense performance.

The approach effectively counters a wide range of adversarial perturbations.

Abstract

This paper introduces a new synthesis-based defense algorithm for counteracting with a varieties of adversarial attacks developed for challenging the performance of the cutting-edge speech-to-text transcription systems. Our algorithm implements a Sobolev-based GAN and proposes a novel regularizer for effectively controlling over the functionality of the entire generative model, particularly the discriminator network during training. Our achieved results upon carrying out numerous experiments on the victim DeepSpeech, Kaldi, and Lingvo speech transcription systems corroborate the remarkable performance of our defense approach against a comprehensive range of targeted and non-targeted adversarial attacks.