At Omsk State Technical University (OmSTU), an innovative neural network has been developed for robust personal data protection. This system utilizes voice authentication, capable of recognizing users even when their voice timbre and intonations change due to emotional states. The findings of this research have been published in the journal Applied System Innovation.
The problem of data breaches is becoming increasingly critical: in the first quarter of 2025 alone, Russian companies faced over 800 million hacker attacks, equating to hundreds of attempts per second. Modern cybercriminals target not only financial and personal data but also sensitive medical and biometric information.
To enhance the security of such sensitive data, OmSTU scientists have created an advanced voice authentication system based on a new neural network model. Pavel Lozhnikov, OmSTU`s Vice-Rector for Research and Innovation, explained that the new algorithm, thanks to its unique neuron types and their mathematical interconnections, possesses increased resilience to external interference attempts.
The developers accounted for various human vocal states—from normal to sleepy, nervous, or tired. During the neural network`s training, datasets included speakers uttering passphrase phrases not only in their normal state but also in altered emotional states.
OmSTU emphasized that their scientific school, specializing in «Secure Artificial Intelligence Neural Network Algorithms,» aims to create solutions that render the leakage or extraction of confidential data from trained AI models either impossible or an extremely difficult and lengthy computational process. The primary issues addressed by this model include improving the accuracy of voice recognition and ensuring the confidentiality of biometric templates from attackers.
In the future, OmSTU scientists plan to adapt this proposed model for other biometric traits, such as handwriting and facial features. Experts also anticipate a rise in attacks on biometric systems using fakes, especially with the advancement of generative artificial intelligence, and are conducting further research to counteract such unauthorized access to information.
