New GATAS method generates adversarial inputs for ASR systems

Researchers have developed a novel black-box testing method called GATAS for automated speech recognition (ASR) systems. This approach generates adversarial inputs by manipulating the latent space of a text-to-speech model, aiming to induce transcription errors while maintaining the naturalness of speech. GATAS formulates the attack as a multi-objective optimization problem, balancing semantic divergence with perceptual quality. Empirical evaluations show GATAS achieves a high success rate with lower distortion and better perceptual quality compared to existing methods, even without direct access to model internals. AI

IMPACT This research introduces a more effective method for testing the robustness of ASR systems against adversarial attacks, potentially leading to more secure and reliable speech recognition technologies.

RANK_REASON The cluster contains an academic paper detailing a new method for testing AI systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Automated Speech Recognition Systems
• Hugging Face
• Text-to-speech modeling
• Transformer-based Models

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →