APPLICATION OF PRE-TRAINED VOSK MODEL IN FACILITY ACCESS CONTROL SYSTEM BASED ON VOICE PASSWORD

Abstract

In today's digital society, voice recognition is rapidly becoming one of the key tools for authentication and access control. The rapid development of artificial intelligence, neural networks, and mobile devices contributes to the introduction of innovative methods that combine a high level of security with maximum convenience for users. Among them, voice authentication occupies a special place - a technology that allows you to identify a person based on their unique vocal characteristics. The article examines the characteristics of access control technologies, as a result of which it was found that modern access control systems based on voice passwords have: vulnerabilities to replay attacks, an average level of security, and require periodic updates. In this regard, voice authentication is a kind of compromise between ease of use and reliability. An algorithm for the operation of an object access control system based on voice passwords has been developed, which clearly separates software and hardware responsibilities, makes the solution fault-tolerant and scalable. A structural diagram of a voice password-based access control system for an object has been developed, which combines software flexibility, hardware minimalism, and high response speed, creating a reliable and affordable voice authentication-based access control mechanism that does not require complex infrastructure.

This approach does not require a permanent connection to the Internet, since the entire recognition process occurs locally on a PC, and the user's biometric data is not stored in the form of audio recordings or other formats, which greatly simplifies compliance with personal data protection requirements. Architecturally, the system is built so that each element performs its own clearly defined set of functions: the microcontroller is responsible only for receiving and processing commands, and the computer part takes on the entire volume of calculations and working with the Vosk voice model. This allows the solution to be easily scaled - if there is a need to connect several access points, it is enough to organize the appropriate number of Python script instances on different computers or mini-PCs and connect them to the necessary control boards. The obtained research results can be used to develop an access control system to an object based on a voice password.