Search academic texts

Information × Registration Number 0226U002773, (0124U000407) , R & D reports Title Research of the resistance of biometric authentication systems to attacks using voice cloning technology based on deep neural networks popup.stage_title Експериментальний та теоретичний аналізи отриманих результатів. Head Opirskyi Ivan R., Доктор технічних наук Registration Date 05-03-2026 Organization Lviv Polytechnic National University popup.description1 The project aims to improve the reliability and security of modern biometric voice authentication systems against attacks using voice cloning technologies. This will allow the development of strategies and protection measures to prevent possible threats in such systems. popup.description2 Within the framework of the research project, a comprehensive theoretical and experimental analysis of the resilience of biometric voice authentication systems to attacks based on modern deep neural network voice cloning technologies has been conducted. A comparative scalability study of ECAPA-TDNN, Pyannote, WavLM base-sv, and WavLM base-plus-sv architectures was performed with the number of users increasing from 10 to 70 on a multilingual dataset. ECAPA-TDNN demonstrated the best balance between accuracy (EER 1.71%) and inference speed (69.43 ms). A patented approach (Ukrainian utility model patent No. 161220) to speaker verification was implemented, based on averaging ten voice embeddings during enrollment and applying dynamic threshold calibration. This method reduced the impact of natural voice variability by 34% and outperformed classical approaches. A specialized dataset of over 3,800 authentic and synthetic voice samples structured into six attack categories (zero-shot, few-shot RVC, voice conversion, replay, hybrid methods) was created. The most critical attack vector was identified as few-shot RVC. A formalized quantitative risk assessment methodology was developed, and a liveness detection module based on nonlinear spectral-phase characteristics was introduced. Product Description popup.authors Dmytro Sabodashko Khrystyna Ruda Yurii Khoma Mariia Shved Anastasiia Zhuravchak Roman Banakh Alina Akhmedova Nazarii Dzhaliuk popup.nrat_date 2026-03-05 Close