mlsanityData to 7 Oct 2026

Multi-Discriminator Sobolev Defense-GAN Against Adversarial Attacks for End-to-End Speech Systems

arXiv 2103.08086 ↗First posted 15 Mar 2021cs.SD cs.LG eess.AScited by 4

Not on the map. Why?

Abstract (arXiv)

This paper introduces a defense approach against end-to-end adversarial attacks developed for cutting-edge speech-to-text systems. The proposed defense algorithm has four major steps. First, we represent speech signals with 2D spectrograms using the short-time Fourier transform. Second, we iteratively find a safe vector using a spectrogram subspace projection operation. This operation minimizes the chordal distance adjustment between spectrograms with an additional regularization term. Third, we synthesize a spectrogram with such a safe vector using a novel GAN architecture trained with Sobolev integral probability metric. To improve the model's performance in terms of stability and the total number of learned modes, we impose an additional constraint on the generator network. Finally, we reconstruct the signal from the synthesized spectrogram and the Griffin-Lim phase approximation technique. We evaluate the proposed defense approach against six strong white and black-box adversarial attacks benchmarked on DeepSpeech, Kaldi, and Lingvo models. Our experimental results show that our algorithm outperforms other state-of-the-art defense algorithms both in terms of accuracy and signal quality.

Key facts. Cited by 4 papers in the map (data through 7 Oct 2026). Builds on 35 of its 64 references that are in the map. First posted on arXiv 15 Mar 2021.

Builds on

35 of 64 references.

References · 35

  1. 2018Large Scale GAN Training for High Fidelity Natural Image Synthesiscited by 2,609
  2. 2017Improved Training of Wasserstein GANscited by 2,223
  3. 2015Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networkscited by 2,048
  4. 2018Spectral Normalization for Generative Adversarial Networkscited by 1,618
  5. 2018Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examplescited by 1,259
  6. 2013Speech Recognition with Deep Recurrent Neural Networkscited by 666
  7. 2017Towards Principled Methods for Training Generative Adversarial Networkscited by 460
  8. 2018Defense-GAN: Protecting Classifiers Against Adversarial Attacks Using Generative Modelscited by 362
  9. 2019MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesiscited by 351
  10. 2018Characterizing Adversarial Subspaces Using Local Intrinsic Dimensionalitycited by 280
  11. 2018Audio Adversarial Examples: Targeted Attacks on Speech-to-Textcited by 268
  12. 2017MMD GAN: Towards Deeper Understanding of Moment Matching Networkcited by 251
23 more
  1. 2015Training generative neural networks via Maximum Mean Discrepancy optimizationcited by 149
  2. 2015End-to-End Attention-based Large Vocabulary Speech Recognitioncited by 148
  3. 2019Imperceptible, Robust, and Targeted Adversarial Examples for Automatic Speech Recognitioncited by 123
  4. 2017The Cramer Distance as a Solution to Biased Wasserstein Gradientscited by 116
  5. 2016Neural Photo Editing with Introspective Adversarial Networkscited by 109
  6. 2019Unsupervised speech representation learning using WaveNet autoencoderscited by 90
  7. 2017Keeping the Bad Guys Out: Protecting and Vaccinating Deep Learning with JPEG Compressioncited by 86
  8. 2019Lingvo: a Modular and Scalable Framework for Sequence-to-Sequence Modelingcited by 86
  9. 2018Robust Audio Adversarial Example for a Physical Attackcited by 42
  10. 2018Targeted Adversarial Examples for Black Box Audio Systemscited by 41
  11. 2019Practical Hidden Voice Attacks against Speech and Speaker Recognition Systemscited by 34
  12. 2017Sobolev GANcited by 29
  13. 2017Learning to Draw Samples with Amortized Stein Variational Gradient Descentcited by 26
  14. 2017Fisher GANcited by 22
  15. 2017On the Effectiveness of Least Squares Generative Adversarial Networkscited by 18
  16. 2019Imperio: Robust Over-the-Air Adversarial Examples for Automatic Speech Recognition Systemscited by 16
  17. 2019Perceptual Based Adversarial Audio Attackscited by 13
  18. 2018ADAGIO: Interactive Experimentation with Adversarial Attack and Defense for Audiocited by 12
  19. 2019A Robust Approach for Securing Audio Classification Against Adversarial Attackscited by 12
  20. 2018Adversarial Machine Learning And Speech Emotion Recognition: Utilizing Generative Adversarial Networks For Robustnesscited by 9
  21. 2019Detection of Adversarial Attacks and Characterization of Adversarial Subspacecited by 7
  22. 2019Deep Griffin-Lim Iterationcited by 5
  23. 2020Class-Conditional Defense GAN Against End-to-End Speech Attackscited by 5

Built on by

Cited by year

Citing papers per year: 2021: 1; 2022: 1; 2023: 0; 2024: 0; 2025: 2.

4 papers.

  1. 2021Bi-Discriminator Class-Conditional Tabular GANcited by 3
  2. 2025Quaternion-Hadamard Network: A Novel Defense Against Adversarial Attacks with a New Datasetcited by 1
  3. 2022RSD-GAN: Regularized Sobolev Defense GAN Against Speech-to-Text Adversarial Attackscited by 0
  4. 2025A Weak Signal Learning Dataset and Its Baseline Methodcited by 0