0:00:15without someone so the world that i'm going to present so you today is and
0:00:20extend personal for one of the sub subsystems so that the ice quality man
0:00:25submitted form needs the elderly and the false and fifteen challenge
0:00:30and the
0:00:32although it is focused on delivery a fifteen we believe that the key components or
0:00:39in this of this paper they can be used in a much wider context
0:00:46so the first thing that we explore is that the how to more get the
0:00:52most of the key lda for
0:00:55and discriminant for discriminative training
0:00:58because nist ovaries that close the set identification task and it's going to train it
0:01:04should be the best
0:01:05so the most important thing is that we show that if we if we're
0:01:11and i u
0:01:12if we used plp parameters to projects i-vectors on the p lda latent subspace apply
0:01:17it discriminative methods in that subspace and then project them but
0:01:21then a week and improve the performance compared to just the baseline the one
0:01:26we use lda and then
0:01:28and maximum mutual information of their from on top of it
0:01:34and the second the
0:01:38important thing is that the we show how four
0:01:45cost function
0:01:47approximated and use it as an objective function for a discriminative l the difference
0:01:52so that as a can see
0:01:55it is based on false acceptance rate and false rejection rate and all of them
0:01:59use indicator functions so we approximate those functions
0:02:04into continues functions and then
0:02:06where able to differentiate them
0:02:11and the of course of these method in general you can be used for any
0:02:16in a cost function in theory
0:02:21okay thank you for attention to the