Audio Adversarial Examples: Targeted Attacks on Speech-to-Text

2018/01/05 by Nicholas Carlini, David Wagner, Carlini, Nicholas +1 · 5 voices · 24 citations
#cs.LG #cs.AI #cs.CR

paper · pdf · doi:10.48550/arxiv.1801.01944

Abstract

We construct targeted audio adversarial examples on automatic speech recognition. Given any audio waveform, we can produce another that is over 99.9% similar, but transcribes as any phrase we choose (recognizing up to 50 characters per second of audio). We apply our white-box iterative optimization-based attack to Mozilla's implementation DeepSpeech end-to-end, and show it has a 100% success rate. The feasibility of this attack introduce a new domain to study adversarial examples.

Cited by

• AdvNav: Behavior-Guided Black-Box Adversarial Attacks on Vision-Language Navigation
• Precision-Varying Prediction (PVP): Robustifying ASR systems against adversarial attacks
• Multi-Level Privacy-Preserving Dementia Detection from Speech via Targeted Adversarial Obfuscation and Representation Learning
• Hijacking Large Audio-Language Models via Context-Agnostic and Imperceptible Auditory Prompt Injection
• Degrading Voice: A Comprehensive Overview of Robust Voice Conversion Through Input Manipulation
• Shedding Light on VLN Robustness: A Black-box Framework for Indoor Lighting-based Adversarial Attack
• Prosody-driven Jailbreaks in Audio LLMs: A Controlled Study and Mechanistic Analysis
• Toward Understanding the Transferability of Adversarial Suffixes in Large Language Models
• Mirage Fools the Ear, Mute Hides the Truth: Precise Targeted Adversarial Attacks on Polyphonic Sound Event Detection Systems
• Evaluating the Robustness of a Production Malware Detection System to Transferable Adversarial Attacks
• When Silence Matters: The Impact of Irrelevant Audio on Text Reasoning in Large Audio-Language Models
• Decoding Deception: Understanding Automatic Speech Recognition Vulnerabilities in Evasion and Poisoning Attacks
• Are Modern Speech Enhancement Systems Vulnerable to Adversarial Attacks?
• Piggybacking on Perception: Stealthy Concurrent Audio Prompt Injections against Multimodal LLM Agents
• SyncBreaker:Stage-Aware Multimodal Adversarial Attacks on Audio-Driven Talking Head Generation
• Impact of Phonetics on Speaker Identity in Adversarial Voice Attack
• ENJ: Optimizing Noise with Genetic Algorithms to Jailbreak LSMs
• Towards stable AI systems for Evaluating Arabic Pronunciations
• When Audio and Text Disagree: Revealing Text Bias in Large Audio-Language Models
• Hallucinations in medical devices
• Are Inherently Interpretable Models More Robust? A Study In Music Emotion Recognition
• When Good Sounds Go Adversarial: Jailbreaking Audio-Language Models with Benign Inputs
• Adversarial machine learning [wikipedia]
• AI safety [wikipedia]

Discussions

• Audio Adversarial Examples: Targeted Attacks on Speech-To-Text [hn, 64 points, 17 comments]
• Targeted Attacks on Speech-to-Text [pdf] [hn, 4 points, 1 comments]
• Audio Adversarial Examples: Targeted Attacks on Speech-To-Text [hn, 3 points, 0 comments]
• Carlini & Wagner, "Audio Adversarial Examples: Targeted Attacks on Speech-to-Text" (Jan 2018). Researchers manage to beat DeepSpeech's speech-to-text engine with adversarial sound w/ 100% success rate [bsky, 0 points, 0 comments]
• Adversarial examples against machine learning systems are such an incredible exploit, because they weaponize the gap in our understanding of those systems. This new one targets speech-to-text extremel [bsky, 0 points, 1 comments]

Related

• Audio Adversarial Examples: Attacks Using Vocal Masks
• Towards Robust Speech-to-Text Adversarial Attack
• WaveGuard: Understanding and Mitigating Audio Adversarial Examples
• Isolated and Ensemble Audio Preprocessing Methods for Detecting Adversarial Examples against Automatic Speech Recognition
• Did you hear that? Adversarial Examples Against Automatic Speech Recognition