MultiPAD: A Multivariant Partition-Based Method for Audio Adversarial Examples Detection

Guo, Qingli; Ye, Jing; Hu, Yu; Zhang, Guohe; Li, Xiaowei; Li, Huawei

doi:10.1109/access.2020.2985231

“…Target Generality Knowledge [8], [15], [24], [38], [43], [63], [64], [73], [120], [124] Before Sensor Universal None [7], [31], [36], [45], [49], [61], [87], [88], [93], [103], [110], [140], [141], [145], [147] Between Sensor and ASR Specific Partial [105], [106], [122], [123], [133] Inside ASR Specific Full perturbation cancellation [36], [45], [49], [110], adding distortion [61], [87], [103], signal smoothing [45], audio compression [31], [88], [147]) to destruct the adversarial perturbation (if any) to protect ASR systems. Other works apply an extra detection network [7], [46], [93], [140], [141] or multi-model detection mechanism [145]. However, those defense s...…”

Section: Defensementioning

confidence: 99%

SoK: A Modularized Approach to Study the Security of Automatic Speech Recognition Systems

Chen¹,

Zhang²,

Yuan³

et al. 2021

Preprint

View full text Add to dashboard Cite

With the wide use of Automatic Speech Recognition (ASR) in applications such as human machine interaction, simultaneous interpretation, audio transcription, etc., its security protection becomes increasingly important. Although recent studies have brought to light the weaknesses of popular ASR systems that enable out-of-band signal attack, adversarial attack, etc., and further proposed various remedies (signal smoothing, adversarial training, etc.), a systematic understanding of ASR security (both attacks and defenses) is still missing, especially on how realistic such threats are and how general existing protection could be. In this paper, we present our systematization of knowledge for ASR security and provide a comprehensive taxonomy for existing work based on a modularized workflow. More importantly, we align the research in this domain with that on security in Image Recognition System (IRS), which has been extensively studied, using the domain knowledge in the latter to help understand where we stand in the former. Generally, both IRS and ASR are perceptual systems. Their similarities allow us to systematically study existing literature in ASR security based on the spectrum of attacks and defense solutions proposed for IRS, and pinpoint the directions of more advanced attacks and the directions potentially leading to more effective protection in ASR. In contrast, their differences, especially the complexity of ASR compared with IRS, help us learn unique challenges and opportunities in ASR security. Particularly, our experimental study shows that transfer attacks across ASR models is feasible, even in the absence of knowledge about models (even their types) and training data.

show abstract

A Multimodal, Multimedia Point-of-Care Deep Learning Framework for COVID-19 Diagnosis

Rahman

¹

,

Hossain

²

,

Alrajeh

³

et al. 2021

ACM Trans. Multimedia Comput. Commun. Appl.

View full text Add to dashboard Cite

In this article, we share our experiences in designing and developing a suite of deep neural network–(DNN) based COVID-19 case detection and recognition framework. Existing pathological tests such as RT-PCR-based pathogen RNA detection from nasal swabbing seem to display low detection rates during the early stages of virus contraction. Moreover, the reliance on a few overburdened laboratories based around an epicenter capable of supplying large numbers of RT-PCR tests makes this testing method non-scalable when the rate of infections is high. Similarly, finding an effective drug or vaccine with which to combat COVID-19 requires a long time and many clinical trials. The development of pathological COVID-19 tests is hindered by shortages in the supply chain of chemical reagents necessary for testing on a large scale. This diminishes the speed of diagnosis and the ability to filter out COVID-19 positive patients from uninfected patients on a national level. Existing research has shown that DNN has been successful in identifying COVID-19 from radiological media such as CT scans and X-ray images, audio media such as cough sounds, optical coherence tomography to identify conjunctivitis and pink eye symptoms on the ocular surface, body temperature measurement using smartphone fingerprint sensors or thermal cameras, the use of live facial detection to identify safe social distancing practices from camera images, and face mask detection from camera images. We also investigate the utility of federated learning in diagnosis cases where private data can be trained via edge learning. These point-of-care modalities can be integrated with DNN-based RT-PCR laboratory test results to assimilate multiple modalities of COVID-19 detection and thereby provide more dimensions of diagnosis. Finally, we will present our initial test results, which are encouraging.

show abstract

SoK: A Modularized Approach to Study the Security of Automatic Speech Recognition Systems

Chen

¹

,

Zhang

²

,

Yuan

³

et al. 2022

ACM Trans. Priv. Secur.

View full text Add to dashboard Cite

With the wide use of Automatic Speech Recognition (ASR) in applications such as human machine interaction, simultaneous interpretation, audio transcription, etc., its security protection becomes increasingly important. Although recent studies have brought to light the weaknesses of popular ASR systems that enable out-of-band signal attack, adversarial attack, etc., and further proposed various remedies (signal smoothing, adversarial training, etc.), a systematic understanding of ASR security (both attacks and defenses) is still missing, especially on how realistic such threats are and how general existing protection could be. In this paper, we present our systematization of knowledge for ASR security and provide a comprehensive taxonomy for existing work based on a modularized workflow. More importantly, we align the research in this domain with that on security in Image Recognition System (IRS), which has been extensively studied, using the domain knowledge in the latter to help understand where we stand in the former. Generally, both IRS and ASR are perceptual systems. Their similarities allow us to systematically study existing literature in ASR security based on the spectrum of attacks and defense solutions proposed for IRS, and pinpoint the directions of more advanced attacks and the directions potentially leading to more effective protection in ASR. In contrast, their differences, especially the complexity of ASR compared with IRS, help us learn unique challenges and opportunities in ASR security. Particularly, our experimental study shows that transfer attacks across ASR models is feasible, even in the absence of knowledge about models (even their types) and training data.

show abstract

MultiPAD: A Multivariant Partition-Based Method for Audio Adversarial Examples Detection

Cited by 8 publications

References 25 publications

SoK: A Modularized Approach to Study the Security of Automatic Speech Recognition Systems

SoK: A Modularized Approach to Study the Security of Automatic Speech Recognition Systems

A Multimodal, Multimedia Point-of-Care Deep Learning Framework for COVID-19 Diagnosis

SoK: A Modularized Approach to Study the Security of Automatic Speech Recognition Systems

Contact Info

Product

Resources

About