SPEAKER VERIFICATION Russian patent published in 2019 - IPC G10L17/02 

FIELD: physics.

SUBSTANCE: invention relates to verification of speaker. Disclosed are methods, systems, a device including computer programs encoded on a computer data medium to facilitate language-independent verification of the speaker. In one aspect, the method includes receiving, user device, audio data representative of a user's speech fragment. Other actions may include input to the neural network stored on the user device, obtained from audio data and a language identifier. Neural network can be trained using speech data representing speech in different languages or dialects. Method may include additional generation, based on the output of the neural network, presentation of the speaker and determination, based on the presentation of the speaker and the second representation, that the speech fragment is a fragment of the speech of the user. Method can provide a user with access to a user device based on determining that the speech fragment is a speech fragment of the user.

EFFECT: technical result is providing accurate verification of authenticity of speakers who speak different languages or dialects.

14 cl, 5 dwg