how can open them
everyone
the paper
i would like to peace and is in type of
text dependent speaker verification system in the did you have a communication channel from is
to do fourteen for comb research syllable
here i show you the all night of this representation improvement you overview of the
paper and four by the hit you have communication introduction and i will show you
the biometric assistance of for this we hit you communication
speaker verification systems
the name
i will keep the performance evaluation flow by
conclusions
firstly
for the task of these research projects that is pure a and biometric systems that
recently that you of communication for building she the revision control
and
this means that you have to the means benny a high frequency but duration communication
channel
so
the main device and phone these the usage of communication channel is what keep okay
this for kentucky is actually use the embedding issue communication so this approach that is
a focus on realistic communication
then
for the navigation control phone the for the authentication of the speaker so
especially for the c must the when this
go into
sub-turn seaport people
in the control pendant this one though full is
is in this one and
sun
some people register of this
of the nice the presence of
so
only a tonight the person speaking
can trying to see the two and the
the sufficiency part
so is the
point been enforcing to set up these the
systems of one the project
and but the problem phone be so the
the average of communication speaker verification is that
this is a
speaker verification system we hasn't but initial durations
and this is short duration
maybe about how one second per second
and up to
chi seconds and
so that's compared to the conventional duration like
one meeting put leading up or
ten seconds and
i usually use the in this as i
alright is a quite short so
we may
focus on this
up opens the by sun solutions and
under the age of
communication in all database
hasn't many problems and
and i goes all you
those of problems and these the
of phone this a
speaker verification
and
and
so we see some solutions of by using
pass phrase
a pass phrases
what three screens the
so we also collect
some proper database
i use the
two
improving in the past and those are all the speech data
verification systems
well also applied the
marty system combination
to form a improvement
the performance of the systems
no we go to
so you
a few hedge of communication
power
so in this
this a finger
so you
bussing that application of the usage of speaker communication so you can see that too
one is the
from the user as it
like six must the u s c n and
so the other part isn't the control not in a purely so this person is
to pass in communication
so we had you have to devise a
unlike what exactly device
and that use the first three seven and
thank initiate quality to the control synthesis and the control centres we applied the
by we present
for the c must and then so at this moment that's it must the so
speak tune the
what we talked is that
with his the name
speech is that
and this piece you transferred to look control panel and the control site also input
these the
speech is to speaker verification and use them for verification
so at
for example and the same time and the console on the also can beep as
a banana
speech is that
like
we present for certificate the identity
numbers of four
for
verification and we also
combined is to the netting and the idea is to can a
two
to improve this the verification performance
now
for
for
speaker verification
proposal
the nine hundred
correct that is because of the usage of speech you that are
alright
as shown here
facility this is a
j of communications speech you
has quite noisy
because i
i is recorded in
in on what development there is a in this environment this noise in baby c
and d noisy this the quite strong
another problem is that the is the a bunch and the
for verification
the open channel the means
then the channel probability can be norm
by
the speaker verification systems
so that so is the quite
ugh
so for this case the
we of course
quite big problems are for channel compensation
so we cannot use the question a
channel compensation
then enclosed
two
but we use the channel mismatch effects for example you cannot use jfa it can
channel factors the or even we cannot use appear at the a
channel factors a
for this the proposed so it is a ha
how difficult t
for this the project and not know why is that not be friend not those
and speech
speech is speech
that means the
during the
you don't then
and
yes speech you is recorded in of these development
so and is obviously but and
is the
up i
is apply applied
why
quite
that you element so
for on the one a test
environment that is the
in a six
so maybe there is this engine
so we sent is the so now engine
the speaker we have speech may be louder than in
in all these development
so also
well that's because speak to now maybe
this speak
speech is speaking we have be plastic
so not a
problem is that the channel frequency and imitation
with the usage of one we had to have a guy so you
and this whole spectrum
range that
for comparison
the first one
it's normal recording without you had you have
communication
and this one is the
recall that with we had you of china
so you can see
the high-frequency part is a sub present match and
and we know for speaker verification
the major
speaker features a
is in the high-frequency part so if this the information is not large
much so maybe
this is a speaker
but if based on performance the we have dropped and whatnot
known disco to the by energy
since the introduction
in this is systems
a bus or you know
all pass phrase based those speaker verification
systems the
this is the input to the g
subsystems a
with the
gmm-ubm but there is a twenty conversion to
systems the jfa and i-vector
because they a
gmms you audios and so they're having many problem and planted has a can be
shared each other so for example as a
the cash
generally
but ubm parameters a and they can share the supply sense that it occurs a
so
so on the proposed systems that
the computation complexity we have be drawn and table two is just reading
so we
with sony's one and then entice systems the is actually the fusion of the
cheese expensive
the fusion
calibration parameters the and the big
can be
changed by using but you a set of development database
and then finally we
with what we get
this goes from the combination of the
g systems
and then he we so
you
the pass phrase and three screens the
and is the verification
personally
for pass phrase and watering knitting is a
what each pass of phase the
of a speaker
we are here the
is the corresponding models and
for the modelling so
a certain that there are k plus phrases the for speaker i and then
we are
you're k plus place
model was and for this because the
so if speaker
say
for one to crying
to be as the speaker i and
with this the
pass phrase and all so we will
if this and autoseek ha
i
and all up to you compare although with the all these utterances all j at
all
and finally we get
that verification
scores no
we so
the database the
clustering phone this
point is if you had you have communication speaker verification
projects and
this database is it was still for parameter changing
presenter's the
they are used the
for ubm training and values for symmetry total variability in the tree i
in i-vector systems chaining
and also used for plp a chaining and i either used for
i can
eigenvoice the fact the eigenvoice the metric chanting
so
one this database and now from different
you minimum and
from different recording bayesian
presenter's a they can
in office environment and visit you had to have china
and so with different distances
between the recording
and receiving
and then we also collect son database and forum
by using d that setting all recordings
you obviously you've elements of for example
is i as are
pending for clean and also we
because you on what you're element is it to simulate the
no real reason is because a development set up for communication
speech
second speech database the recording with the we had you have
and here is the recording devices
like what we talking
mike and on the microphones and also i pay that is the mobile phone
and
so we have development
a real time
systems the phone this approach that's
i think you know we so
know how well components of the voice the biometric systems the how about
improve the computer
and
this usb song call
and
this of you had you have
has said that there is a walkie talkie here
for receiving and also for just meeting
and here we so the so well user interface and in this the survey into
of
user interface and not cheap regions
the first case i is used for any stray showing up registrations
and then
the second one is a for enrollment
pigeons the and the so one is for test patch pigeons and so
this has being updating find the
by using the idea is to go
test inside
on what
now we go to the performance evaluation
so we can see
and in this the
the pass phrases the
for the evaluation proposal
you know
one this purpose of the participate the we s p
one by one the name
the i b
and
no but repeat several times it
in different sets when i in different development with samples the in different should
so
so here we so
than the
the evaluation database and the development
when database it
the number of what goes the we use the phone this the performance evaluation
and the number of chaining and has a
utterances used of one these evaluation
also we so the true trier the number of trying to try
number of impostor trials
use the phone the evaluation
and we separate the
and not twenty speakers
participating for these the database that
recordings the and we separate and check
this is because the and
and ten speakers and
i'll four
one evaluation and for development purposes
and here
we also given that
the averaging durations the
for the name and for the i d
for
and also for next bus i d
and you can see the averaging
duration is about one point joe four
for the name i is the one point g six that
for all i e
and one then pass i the it can reach
two point four seconds
so
we so
the performance
these are in terms of eer and minimum dcf
for each of
the single system is
and the fusion designs and
and you can see
for things and disaster is always better than the single system the gmm
and jfa and i-vector
why i like to is the performed
is not so cool
as compared to add rosa
so actually
because the in i characteristics than the
we only encode to the
those the
channel information as aforesaid pen
in reality a this there's a
but these channel compensation
"'cause" that is right isn't
what is "'cause" consideration is not so single
for this is duration so
so maybe i you we all make
then the pierrette the a performance draw time
so
in ten so meeting mindcf we also
so the best performance is all that fusion and
so compare always in the second leading and single id than the implies i
performs a
better
then every
in every
systems a
single or
that there's and one
so here we also
can
the
the fusion
with the name process id
current
alright at
in a ten point one cheaper same but
of eer
so is the
quite good results and
we expect
you know from the second
perform an performance of with so here
with the
det plots for you had you have
then
i t and
then they about i the comparisons
so we can see
these the things and results the opportunity better than and the other subsystems the
for name for i b phone n-grams id
and also can see
banana
i e
then the performance that is quite good
now we go to
the conclusion on this the presentation we haven't introduces a
we have introduced a possibly bayes the text dependent speaker verification system
against the industrial
duration condition
we have
develop appears as is then consisting of gmm ubm jfa and i-vector
among then
the ubm and the stuff reasons they do they got astra
and according to the different conditions that between enrollment and but indication we
besides the suitable in these four
for parameter changing and find us
system setup
experimental results or that there's insisting gives the of one system or what any single
system
then
two point four second duration or like eer of that's and then chapter seven
this is my presentation sink
so
for this application i assume
and correct me if i'm wrong where is your operating space i assume that for
the most part if boats are coming in the most the time it's expected that
the right person is gonna be radio in
so you really care about so my correct and that you really care about the
the very low miss rate
is that correct you basically you care what region are you most such an extremely
low miss rate
we just terrible
the identity of a person's
right jet set operating point
right so for this scenario laid out boats coming in this in general so that
it is sense
i guess and try to get a sense like here and i think even a
lot of the text and then applications people are talking about where kind of the
low road
we're focusing on a different part of the debt curve that we would when it's
trying to find a low prior target in the dataset this actually maybe in the
hyper prior target the cost of changing a lot so
do you care what region you gave like equal error rates all that you have
an idea where you really care about operating the system where it's gonna make its
threshold
so
we may consider okay
we go to this
this
for this part
then
we can see
for this one
we may consider to use the automatic speech recognition machine use it to replace in
this control on the so that means that we can improve the total performance of
less is then so that
by total automatically
sue
zero in this the
automatically by or not
systems that and get
then the then the information of the
us because the and to the verification when this task as
then this
idea can improve in the total
performance
the
sorry
in some way
i'm interested in the communication part of fuel system you talk is entitled v h
if communication
we had to have communication is not very specific it simply means that the radio
frequency ranges between city and three hundred megahertz but there are many ways and many
different channel qualities and signal quality set you can transmit over v h if so
i think you implied that you use marine radial which is the
usually analogue and f m but not necessarily you can transmit the signal digitally
in many different modulation the channels and then i assume that you talking just about
the marine the walkie talkies but then in your list of databases you also mentioned
mobile phone data now mobile phone data is not either transmitted on v h f
no analog so i'm confused how you use that data in analysing your the range
channels
so from this that isn't why we choose and those them about mobile phone devices
because the we don't have enough
database to use the
with the did you have
a friend from the system changing so we haven't tried several times it
by we use
by discussing this ball database of button the performance of we have dropped so
we i in the sun
some that are from this a mobile device and recording
i in many
had of course so this is a
one is a consideration
we only based on
the experimental results
sink
formatting communication whole most cases it we only use the
this we had you have
like walkie talkie for communication
is a popular so is a suitable
for universal
six communication with the control panel
so it normally when you look at ship to ship or maureen type communications in
be modulation demodulation process
quite often in the in b d modulation
if the speech bandwidth is not shifted back to the right location to be an
offset in there
and so that distortion will actually introduce a lot of problem so you have to
kind of a cadre normalization or adjusting here
are you looking at real data when you're doing you're testing and if so what
is be the plan to kind of interest some of the other
problems of used to be christian analogue
v h after i
communications because i don't see you have listed