Аннотация
In the digital age, the importance of biometric identification technologies is increasing due to the increasing demand for reliable, secure, and user-friendly access control systems for electronic devices. One of the promising areas of such technologies is speaker authentication, which allows confirming identity based on a person’s unique voice characteristics. However, traditional methods cannot provide the required level of stability in different acoustic environments, background noise conditions, and changes in speech features. In the presented research work, a comprehensive approach is proposed that combines multi-frequency coefficients (MFCCs), x-vector-based speaker inputs, and deep neural networks (DNNs) to improve the accuracy and stability of speaker authentication. The proposed architecture consists of preprocessing speech signals, extracting MFCC features, forming x-vector inputs, and classifying speakers using DNN. An experimental dataset consisting of audio recordings of 100 speakers was used to evaluate the effectiveness of the proposed model. As a result of the experiment, the proposed model achieved 98.5% authentication accuracy, 99.4% precision, 97.6% recall, and a 98.5% F1-score. In addition, the false acceptance rate (FAR) was 0.6%, the false rejection rate (FRR) was 2.4%, and the equal error rate (EER) was approximately 1.5%. These results indicate that the model can distinguish between authorized and unauthorized users with high accuracy. The proposed approach showed superior results compared to traditional machine learning and deep learning models, which proved the effectiveness of combining acoustic features and deep speaker embeddings. The results indicate that the proposed method can be used in secure access control to electronic devices, biometric user identification, and voice authentication systems.