0:00:16i or the u s and model
0:00:22a target o k
0:00:26the weighting rule
0:00:29a problem in the unimodal
0:00:35i all right
0:00:40okay
0:00:41i'm in the table
0:00:43a little or it all i mean how men
0:00:49and
0:00:51to improve the performance you know i wrote you might not
0:00:59it better
0:01:00but the is better
0:01:03we are currently are really hartman and recognition
0:01:08okay tomorrow i in rate in one or more
0:01:12i don't okay
0:01:14you map or
0:01:16a little more to the global mean not
0:01:21a more or the u i
0:01:27the problem we will have
0:01:31i
0:01:35we gotta identification
0:01:38and the
0:01:39a reasonable that the of course i
0:01:45although
0:01:46a new
0:01:48notation o a whole number one
0:01:51but i three right okay
0:01:55general model unit
0:01:58i write my
0:02:03the other hand
0:02:05or you might add more efficient model
0:02:11probably not you know there
0:02:19what the motivation for
0:02:22no i nor my usual the problem all the speaker recognition
0:02:29on the future
0:02:30only we use when i
0:02:34the are you speaker
0:02:38nothing better than
0:02:41a dimension
0:02:43well i not information in
0:02:46the core condition
0:02:50really
0:02:52well i
0:02:55the mobile
0:02:57okay well when i don't recognition model
0:03:04the model
0:03:05okay well
0:03:07i
0:03:08convolutional layer it more data
0:03:12the speaker model
0:03:15most people do you know well
0:03:18we more
0:03:21other than i'm recognition model
0:03:24a the fact
0:03:26independently
0:03:27and to get
0:03:31a little better than the one point five k u one and recognition
0:03:40i mean that no the one that
0:03:43it is not who and the in domain and with it
0:03:52well
0:03:53i
0:03:57in all the speaker and you know
0:04:00a whole and out in a
0:04:05if you allusion and what it is you
0:04:08well
0:04:13of the movie
0:04:14a more efficient addition there then you have the time
0:04:21in addition there than l
0:04:23well information open condition
0:04:28you know the
0:04:31the in the morning edition
0:04:34a model adaptation okay really global condition
0:04:38a eight conversation edition
0:04:42and the
0:04:44we have only okay
0:04:49on the other hand recognition
0:04:52i don't
0:04:53a really and i will be a really by a all out of domain
0:05:02well i and you the mac layer generating the n r
0:05:10the calibration they the i
0:05:13a binary okay or more
0:05:18the little i
0:05:21well by the channel estimate
0:05:26on the other habitation
0:05:29other application had available
0:05:34a be a
0:05:35but we directly are really apply channel
0:05:40a where we have we are counting on timit
0:05:47i between the core condition
0:05:52the idea is to a you know not to buy or by two
0:06:01i'm the weight
0:06:03i really
0:06:05using a not very i don't like to buy is
0:06:10a visual word by
0:06:13you the
0:06:14the application i
0:06:17i'm from the speaker who
0:06:20we
0:06:22at a
0:06:23okay the handle because we have i
0:06:33and
0:06:36probably okay
0:06:38the
0:06:42enrollment data about
0:06:46i o
0:06:49i u
0:06:51either the
0:06:52a longer than that
0:06:55where you that you are you
0:07:00e o to o e
0:07:03every meeting recognition
0:07:06although the without a
0:07:10and the
0:07:12i know
0:07:14i don't you do
0:07:17and the other hand but
0:07:21only one done by the okay
0:07:26okay in that
0:07:28well
0:07:29well everyone to you to the
0:07:33a e
0:07:35and okay you did not
0:07:38we have
0:07:39the or
0:07:47i
0:07:49but i
0:07:51the
0:07:51speaker identification
0:07:53you
0:07:55the one approach
0:07:59a that a whole training and that
0:08:04i don't know there is no longer than the other hand but
0:08:10a menu that and y two k
0:08:16okay and that
0:08:17okay
0:08:19it
0:08:20one okay
0:08:23you know to do the whole problem
0:08:27i don't know that will be used to a well known
0:08:32did you are at the not and not you
0:08:37are you that
0:08:40only option
0:08:42we you know
0:08:43i but the i and i'll
0:08:48i for the
0:08:49but the it that
0:08:53and you as a nice improvement
0:08:56well
0:08:57and i can be i
0:09:01and maybe a data
0:09:04a well you all the all pro
0:09:11and there is a okay i'm weight and the
0:09:16or you a bit
0:09:19and i'm time and then the more
0:09:23you have the right
0:09:25and the other p
0:09:31okay
0:09:34and one speaker recognition
0:09:37there are totally
0:09:39really and then
0:09:43okay
0:09:45a fusion of it or not
0:09:50well
0:09:52and then than a the thing but okay not only
0:09:58no value
0:10:00a very
0:10:02and that
0:10:04i just a the university human
0:10:09i'm not known
0:10:14well
0:10:23i
0:10:24that
0:10:25and it will be
0:10:29the i e
0:10:32we then
0:10:33i'm getting higher than i would be other
0:10:38and i three
0:10:39a speaker verification and you the speaker recognition model only
0:10:45i e
0:10:47i e
0:10:48are there
0:10:51data a little
0:10:53and the speaker recognition more
0:10:55a using them
0:11:01i right
0:11:02i e
0:11:04i four or more using a total and one iteration
0:11:09well you
0:11:11and i e
0:11:13okay we have at
0:11:15the i a model that using a total between and among the more you
0:11:22the recognition
0:11:25i'm the of the more like a you know how you know what
0:11:31i am with organic probably layer
0:11:35we don't well
0:11:37i mean all
0:11:39a little the
0:11:41well
0:11:44and a
0:11:45okay the problem of the one level
0:11:50i
0:11:51it might have an error rate
0:11:55a in well i don't really condition
0:11:59and order to make the model i to make a movie that it would be
0:12:06and the television if you
0:12:09a it's a convolutional
0:12:12well
0:12:14okay
0:12:15and then combine them here
0:12:17there are more efficient
0:12:21and the actual the model
0:12:27and do not well what is that it may be a problem
0:12:35well
0:12:37i the them you
0:12:39the we didn't use a suitable
0:12:43but you wanna you know what
0:12:47and that we the problem you
0:12:54and the
0:12:55but there is no
0:12:57the
0:13:00it'll
0:13:01well
0:13:05you the more you rate
0:13:11i e
0:13:13i performance of the extended edition
0:13:17a
0:13:19the united
0:13:25and
0:13:28i
0:13:28i and i
0:13:30and i e and i that the
0:13:33and you want to than what we implement a one
0:13:42a very good that you only not
0:13:47how to i i e
0:13:50and you know data model
0:13:53she'll
0:13:54by a nice improvement
0:13:56you know a little you that'll
0:14:00and i probability a really all you can
0:14:06i
0:14:07the information and okay
0:14:13i
0:14:14you know well
0:14:18and then
0:14:19but the
0:14:22recognition
0:14:23a lot of the
0:14:27no
0:14:28the
0:14:29here
0:14:31but not or you
0:14:35from people i and so that
0:14:39through a piranha
0:14:41the individual model
0:14:47if i the you know little or you can perform i might not
0:14:55right
0:14:56you can we try to model i s o a long
0:15:02having to do not
0:15:05them
0:15:06relative improvement okay
0:15:08in r e i e
0:15:11i e at a
0:15:16but the more
0:15:21for the whole speaker recognition
0:15:25a little i know the true value then problem the model
0:15:33it completely okay on a
0:15:37v you in a further more
0:15:47there you go fishing model you years you model and a like i of it
0:15:55you more normal or
0:15:59then there is a little more i
0:16:05the i
0:16:07only the that
0:16:09the biphone like
0:16:11i don't you
0:16:12okay i
0:16:14you are
0:16:16right
0:16:22you tomorrow you are model
0:16:25my anymore or and then
0:16:29the women
0:16:30information
0:16:31and you have two more years
0:16:35what do you all
0:16:42and the three the original the real identification
0:16:49and the
0:16:51i i'm at
0:16:53but i thirty and i
0:17:00in a file
0:17:03you know the a i e i'm sorry
0:17:08i don't know that or a i u i e
0:17:17of the plan
0:17:21well
0:17:23a mission it you know that only you the a i e and but i
0:17:29e
0:17:30and the barrel
0:17:32the new combination the in narrative on the that
0:17:37at the end and i
0:17:40it is clear from the
0:17:42that are not
0:17:45a final no
0:17:47bigger
0:17:48i you know an additional
0:17:51the video i e
0:17:55if you're gonna i and
0:17:58and i and i
0:18:00a the i know
0:18:04well you
0:18:05i
0:18:06you
0:18:08i
0:18:09the
0:18:11recognition
0:18:12although
0:18:14on the number three
0:18:18a the are really article
0:18:26the level
0:18:28and the by i i'm the colour
0:18:36the p and
0:18:43i didn't
0:18:45on the paper
0:18:47a lot of data by the
0:18:50six hub four
0:18:53and the unit for
0:18:56well
0:18:58you know the two
0:19:00cool
0:19:00the
0:19:02identification
0:19:03and that a simple
0:19:05the or not
0:19:10a possible
0:19:12there is more
0:19:15well you never the
0:19:19all the model
0:19:21and i
0:19:24the relevant information
0:19:27but the problem then you're able to spend the only you the graph model
0:19:34and the over
0:19:36okay
0:19:38i four u o a i-vector model kind
0:19:41what the
0:19:42you great
0:19:48or
0:19:50and future work
0:19:52that is that you the multi condition
0:19:55i can be really you
0:19:59and the what we're very well but
0:20:04a movie that i
0:20:06okay i four u
0:20:10you