To obtain access to full text of journal and articles you must register!
- Article name
- Transformer of voice passwords of speakers into a cryptographic key based on a committee of pretrained convolutional neural networks
- Authors
- Sulavko A. E., , sulavich@mail.ru, Omsk State Technical University, Omsk, Russia
Stadnikov D. G., , sdg250598@inbox.ru, Omsk State Technical University, Omsk, Russia
Choban A. G., , adil_choban@mail.ru, Omsk State Technical University, Omsk, Russia
Inivatov D. P., , sulavich@mail.ru, Omsk State Technical University, Omsk, Russia
- Keywords
- multilayer neural networks / deep learning / ensembles of models / automatic learning of neural networks / speaker voice parameters / auto-encoders / convolution core / biometric authentication
- Year
- 2021 Issue 4 Pages 23 - 33
- Code EDN
- Code DOI
- 10.52190/2073-2600_2021_4_23
- Abstract
- A method has been developed for converting a voice password into a long cryptographic key or a strong password for reliable biometric authentication. A new method of ensemble of classifiers is proposed, which can be used to reduce the number of pattern recognition errors, including in conjunction with methods such as bagging, boosting, and stacking. Five pre-trained multilayer convolutional neural networks are combined into a committee. Each network was trained on the average spectra of voice images calculated using the fast window Fourier transform using various window functions (rectangular, Barlett, Gauss, Blackman, Hamming). The networks extracted voice password feature vectors that were input to the perceptron based converter "biometrics to code", trained according to the GOST R 52633.5 algorithm. The achieved error level was EER = 0.0144.
- Text
- BUY for read the full text of article
- Buy
- 500.00 rub
