DBAC-DSR-BT: A secure and reliable deep speech recognition based-distributed biometric access control scheme over blockchain technology

Abstract

Speech recognition systems have been widely employed in several fields including biometric access control. In such systems, handling sensitive data represents a real threat and risk to security and privacy, namely in the central environment. This paper proposes an innovative solution that integrates speech recognition power as a biometric modality with the decentralized and tamper-resistant nature of blockchain technology aims at designing, implementing, and evaluating an access control system that not only leverages the unique characteristics of speech recognition through the AutoEncoding Generative Adversarial Network (AE-GAN) model for user authentication but also ensures the enforcement of access policies and voice templates storage through two distinct Smart Contracts. The first smart contract aims at storing the ID of encrypted templates matched to the hash of the public address and encrypted attributes. While the second smart contract incorporates the security policy and takes charge of generating an access token if the conditions have been satisfied. Which makes it easier to upgrade specific components without affecting the entire system. Moreover, this architecture delegates the extraction features, conversion into template, encryption, and similarity calculation functions of encrypted templates using homomorphic encryption to an API to provide more security, privacy, scalability and interoperability and reduce the overhead within the blockchain. This API interacts with the smart contract using Oracle services that ensure the interaction between on-chain and off-chain, which provide a reliable, fine-grained, and robust scheme. The simulation of this proposed scheme proves its robustness, efficiency, and performance in terms of security, reliability, and resistance to several attacks.