As the demand for advanced data security intensifies, traditional authentication methods are increasingly rendered inadequate, exposing users to breaches and theft. Voice biometrics, leveraging the unique and immutable vocal characteristics of individuals, has redefined secure authentication by offering scalability, adaptability and seamless user convenience. This research harnesses the potential of the XEUS model, a cutting-edge framework powered by self-supervised learning, to analyse and extract high-dimensional speaker embeddings from the VoxCeleb1 dataset. The XEUS model’s advanced capability to extract speaker-specific acoustic patterns directly from raw data reduces dependency on extensive labelled datasets, making it an ideal choice for diverse and complex authentication scenarios. The study lays the foundation for a sophisticated voice biometric framework that employs phonetic and prosodic analysis leveraging speaker embedding extraction to capture vocal features (Horiguchi et al. in Guided speaker embedding, 2024 [1]). Built on self-supervised learning, the XEUS model autonomously learns speaker representations from raw data without requiring labelled inputs, enhancing its adaptability to real-world, large-scale authentication challenges (Khan et al. in Int J Appl Eng Manag Lett, 2022 [2]).