🤖 AI Summary
To address backdoor vulnerabilities in speech recognition systems, this paper proposes MarketBack—a novel audio backdoor attack method inspired by stochastic investment principles in finance. MarketBack models the random investment process as a dynamic and stealthy trigger mechanism, employs Bayesian optimization to adaptively tune perturbation parameters, and injects imperceptible, style-aware audio perturbations for low-overhead data poisoning (<1% poisoned samples). Unlike conventional static-trigger approaches, MarketBack significantly enhances attack generalizability and cross-model transferability. Extensive experiments across seven mainstream speech models demonstrate near-perfect average attack success rates (~100%), while maintaining perceptual invisibility of the perturbations. This work introduces a new paradigm for audio AI security research and establishes a strong baseline for evaluating robustness against adaptive, behaviorally grounded backdoor attacks.
📝 Abstract
With the growing use of voice-activated systems and speech recognition technologies, the danger of backdoor attacks on audio data has grown significantly. This research looks at a specific type of attack, known as a Stochastic investment-based backdoor attack (MarketBack), in which adversaries strategically manipulate the stylistic properties of audio to fool speech recognition systems. The security and integrity of machine learning models are seriously threatened by backdoor attacks, in order to maintain the reliability of audio applications and systems, the identification of such attacks becomes crucial in the context of audio data. Experimental results demonstrated that MarketBack is feasible to achieve an average attack success rate close to 100% in seven victim models when poisoning less than 1% of the training data.