and i don't have circle
i four u
the speaker recognition workshop
the decision or not
the right
each one
german
furthermore
e
i think
okay what the
or the u i i think and i think
right
in the u
on a greater and would be efficiently
can he may be paid on time nine cepstral coefficients
then you are using it can also report that you can be one or two
and
evaluation and you know
and then just
all right
i wouldn't can control only
i don't
and two
because you
section
finally and b
that is not required to read
right behind the original speaker
i to me
and
i
or not
i
however
it just one and evaluated on the next
this means that
right
and
phonotactic features
all right
why only r e one
it can be or finally we got to each of your equally well in that
one
in
in the gmm be in the u i
vectors
i the only a very
where and how i-vectors
i mean
on
it is
lda and you know
to provide a little time
that can be okay well or better
the main motivation behind
in one
i really
why the into it i think i might not be able to the we were
able
section four
there are derived from a wide open set of the well
in
i
in order
and
then
but development
finally you know
no
we can be different
based in it
the window
in an image
i and i
furthermore i
on the one
but
then the invention limited
i mean vector that f and g
a vector right in one
or what do not all the unit vector
well fourteen
and
i
one
sure that
i computation
the low
it
i
and
and i in
the problem
the
i mean
and all
i
and
wait
like image over time
finally the update
you're right
you date
okay
but i mean not
you anyone e
and one two
don't have been working
i don't want to
and
and only one working
computed
one
okay and option
one can be
and more original have to be a and you don't dimensional
i o compute
you are
i
and
for the computation
and that
it also
okay we have
one computation
i
do not need a one and e
i
i think
i mean and unit one
i
i
and a
one hundred
i
i-vectors
i
well no more than one might be
and one
what
and german
how well thank you
okay
and i
you and all three in
management for evaluation
e
we want you
i
and the way i
from i
sure
the i-vector
and then
and
for every
we can do
but i
two
eight
or not
and
in section
therefore important
for that
you or i
okay
i and i one
and in fact it can work on
in view
one o d baseline
and
what do you
i one
i dunno
or you know how one
and
women
and it can
the quantization
where are in trouble
i'm sure that you can be
one
the p
i
a company
i
okay and
on the gain
and
and england for to be able
and that we propose
you know
i four u
will be data
a big
the point two
in our
okay
in the evaluation data
based on the incline
after that you by the
i don't the problem
cool
all the time domain
in
rule
right
well i
i don't know that in detail
a woman and the right
what i don't
the whole
well
in room b
i don't
it didn't
from the number and can be
the problem
we can write a
that is
our balloon of entropy
my little or no you're enjoying of the can go into a low dimensional
right
in particular or
by really global warming well
i don't think that are more than
and so on
but i that time
and they are
our one
or can be a problem i-vector i four point one high dimensional
for all of them
and the i-vector
the do not
and i can be the most
i don't
of course
for example
in the back one
quite good
equation one point one
and hundred and an i-vector
the only one hundred and d o b i
that was used
have you can be done
okay
and i mean and i think i four
and you
well
that you can be done on a and b
by
one working
i four point four in can perform better than four i-vectors
and i can
i
i that
and what have one
alright
one
when they i and i know how much of the wood
and one time
i one and
where
in performance
i
very well
and that kind of a lot of iraqi
and you