Speech recognition systems driven by Deep Neural Networks (DNNs) have revolutionized human-computer interaction through voice interfaces, which significantly facilitate our daily lives. However, the growing popularity of these systems also raises special concerns on their security, particularly regarding backdoor attacks. A backdoor attack inserts one or more hidden backdoors into a DNN model during its training process, such that it does not affect the model’s performance on benign inputs, but forces the model to produce an adversary-desired output if a specific trigger is present in the model input. Despite the initial success of current audio backdoor attacks, they suffer from the following limitations: (i) Most of them require sufficient knowledge, which limits their widespread adoption. (ii) They are not stealthy enough, thus easy to be detected by humans. (iii) Most of them cannot attack live speech, reducing their practicality. To address these problems, in this paper, we propose FlowMur, a stealthy and practical audio backdoor attack that can be launched with limited knowledge. FlowMur constructs an auxiliary dataset and a surrogate model to augment adversary knowledge. To achieve dynamicity, it formulates trigger generation as an optimization problem and optimizes the trigger over different attachment positions. To enhance stealthiness, we propose an adaptive data poisoning method according to Signal-to-Noise Ratio (SNR). Furthermore, ambient noise is incorporated into the process of trigger generation and data poisoning to make FlowMur robust to ambient noise and improve its practicality. Extensive experiments conducted on two datasets demonstrate that FlowMur achieves high attack performance in both digital and physical settings while remaining resilient to state-of- the-art defenses. In particular, a human study confirms that triggers generated by FlowMur are not easily detected by participants. The source code of FlowMur is publicly available at https://github.com/cristinalan/FlowMur.

Voiceprint authentication is a promising approach for user authentication due to its low acquisition cost and non-contact characteristics. However, most of existing voiceprint authentication systems directly store user voiceprint templates in a database, without taking any effective measures to protect user privacy. Once the voiceprint templates in the database are leaked, user voice information becomes public, which brings severe security threats to the users and impact practical applications of a voiceprint authentication system. Few literature works study privacy-preserving voiceprint authentication although it is particularly needed. Most of existing solutions based on cryptography suffer from high computational costs. But usability of some efficient systems is poor due to additional interactions with users. To tackle the above issues, this paper proposes VoiceSketch, a privacy-preserving voiceprint authentication system based on a secure sketch to protect the privacy of user voiceprint templates with high usability. VoiceSketch does not store the voiceprint templates directly. Instead, it uses a secure sketch to correct errors between multiple voiceprint features of the same user and use cryptographic hash function to extract an authentication key from a user’s voiceprint template. We analyze the security of VoiceSketch based on secure sketch and hash function. We also implement VoiceSketch and design a series of experiments to verify its accuracy and effectiveness.