a lot easier or a improving the robustness of speaker verification systems against maybe speech

then i'll

tested with a

the original voice of the impose that the speaker state-of-the-art speaker verification systems perform better

but that the impersonator is mimicking the target speaker the performance deteriorates significantly

be addressed this issue you know in our work

we use a proximal support vector machine based backend classifier its use i-vectors that input

for improving the robustness against a mimicked speech

then be devised as tight as it do maybe each of the training examples for

this em to further improve the robustness of the system against mimicry it's is the

or stuff that was not but you

thank you