0:00:16and i don't have circle
0:00:24i four u
0:00:26the speaker recognition workshop
0:00:32the decision or not
0:00:37the right
0:00:40each one
0:00:43german
0:00:45furthermore
0:00:47e
0:00:54i think
0:00:56okay what the
0:01:01or the u i i think and i think
0:01:09right
0:01:19in the u
0:01:22on a greater and would be efficiently
0:01:27can he may be paid on time nine cepstral coefficients
0:01:34then you are using it can also report that you can be one or two
0:01:49and
0:01:51evaluation and you know
0:01:55and then just
0:02:00all right
0:02:02i wouldn't can control only
0:02:05i don't
0:02:07and two
0:02:10because you
0:02:12section
0:02:13finally and b
0:02:21that is not required to read
0:02:26right behind the original speaker
0:02:30i to me
0:02:34and
0:02:36i
0:02:38or not
0:02:43i
0:02:48however
0:02:49it just one and evaluated on the next
0:02:55this means that
0:03:10right
0:03:14and
0:03:15phonotactic features
0:03:19all right
0:03:24why only r e one
0:03:31it can be or finally we got to each of your equally well in that
0:03:44one
0:03:46in
0:03:47in the gmm be in the u i
0:03:52vectors
0:03:55i the only a very
0:03:59where and how i-vectors
0:04:04i mean
0:04:08on
0:04:14it is
0:04:15lda and you know
0:04:27to provide a little time
0:04:33that can be okay well or better
0:04:41the main motivation behind
0:04:49in one
0:04:52i really
0:04:56why the into it i think i might not be able to the we were
0:05:10able
0:05:11section four
0:05:14there are derived from a wide open set of the well
0:05:21in
0:05:22i
0:05:24in order
0:05:32and
0:05:38then
0:05:41but development
0:05:48finally you know
0:05:53no
0:05:53we can be different
0:05:59based in it
0:06:06the window
0:06:09in an image
0:06:12i and i
0:06:18furthermore i
0:06:20on the one
0:06:22but
0:06:26then the invention limited
0:06:34i mean vector that f and g
0:06:47a vector right in one
0:06:52or what do not all the unit vector
0:06:56well fourteen
0:06:59and
0:07:00i
0:07:02one
0:07:04sure that
0:07:09i computation
0:07:15the low
0:07:18it
0:07:21i
0:07:23and
0:07:26and i in
0:07:30the problem
0:07:32the
0:07:33i mean
0:07:34and all
0:07:40i
0:07:44and
0:07:45wait
0:07:47like image over time
0:07:50finally the update
0:07:54you're right
0:07:55you date
0:07:59okay
0:08:05but i mean not
0:08:11you anyone e
0:08:15and one two
0:08:17don't have been working
0:08:19i don't want to
0:08:24and
0:08:26and only one working
0:08:28computed
0:08:31one
0:08:34okay and option
0:08:41one can be
0:08:45and more original have to be a and you don't dimensional
0:08:54i o compute
0:08:59you are
0:09:01i
0:09:02and
0:09:04for the computation
0:09:12and that
0:09:14it also
0:09:20okay we have
0:09:24one computation
0:09:30i
0:09:32do not need a one and e
0:09:39i
0:09:42i think
0:09:46i mean and unit one
0:09:50i
0:09:51i
0:09:53and a
0:09:57one hundred
0:09:59i
0:10:00i-vectors
0:10:03i
0:10:10well no more than one might be
0:10:15and one
0:10:24what
0:10:31and german
0:10:34how well thank you
0:10:39okay
0:10:43and i
0:10:44you and all three in
0:10:50management for evaluation
0:10:53e
0:10:56we want you
0:10:59i
0:11:01and the way i
0:11:03from i
0:11:05sure
0:11:09the i-vector
0:11:12and then
0:11:14and
0:11:18for every
0:11:19we can do
0:11:21but i
0:11:23two
0:11:25eight
0:11:27or not
0:11:29and
0:11:31in section
0:11:33therefore important
0:11:37for that
0:11:39you or i
0:11:43okay
0:11:46i and i one
0:11:51and in fact it can work on
0:12:00in view
0:12:03one o d baseline
0:12:07and
0:12:10what do you
0:12:14i one
0:12:19i dunno
0:12:21or you know how one
0:12:25and
0:12:27women
0:12:30and it can
0:12:34the quantization
0:12:37where are in trouble
0:12:43i'm sure that you can be
0:12:48one
0:12:50the p
0:12:51i
0:12:56a company
0:12:58i
0:13:00okay and
0:13:02on the gain
0:13:03and
0:13:04and england for to be able
0:13:12and that we propose
0:13:15you know
0:13:19i four u
0:13:23will be data
0:13:25a big
0:13:26the point two
0:13:34in our
0:13:40okay
0:13:42in the evaluation data
0:13:46based on the incline
0:13:50after that you by the
0:13:56i don't the problem
0:13:59cool
0:14:02all the time domain
0:14:04in
0:14:06rule
0:14:07right
0:14:13well i
0:14:24i don't know that in detail
0:14:27a woman and the right
0:14:41what i don't
0:14:43the whole
0:14:46well
0:14:48in room b
0:14:55i don't
0:15:00it didn't
0:15:05from the number and can be
0:15:08the problem
0:15:13we can write a
0:15:17that is
0:15:18our balloon of entropy
0:15:24my little or no you're enjoying of the can go into a low dimensional
0:15:37right
0:15:38in particular or
0:15:51by really global warming well
0:15:55i don't think that are more than
0:16:05and so on
0:16:10but i that time
0:16:19and they are
0:16:22our one
0:16:25or can be a problem i-vector i four point one high dimensional
0:16:46for all of them
0:16:51and the i-vector
0:16:54the do not
0:16:58and i can be the most
0:17:02i don't
0:17:04of course
0:17:10for example
0:17:12in the back one
0:17:17quite good
0:17:18equation one point one
0:17:27and hundred and an i-vector
0:17:29the only one hundred and d o b i
0:17:43that was used
0:17:47have you can be done
0:17:53okay
0:17:54and i mean and i think i four
0:18:00and you
0:18:01well
0:18:03that you can be done on a and b
0:18:14by
0:18:15one working
0:18:17i four point four in can perform better than four i-vectors
0:18:32and i can
0:18:34i
0:18:37i that
0:18:42and what have one
0:18:47alright
0:18:49one
0:18:56when they i and i know how much of the wood
0:19:08and one time
0:19:15i one and
0:19:19where
0:19:21in performance
0:19:28i
0:19:29very well
0:19:31and that kind of a lot of iraqi
0:19:38and you