have a we all my name on the know
i do right to well the of the initial model control whole
generalized
sidelobe canceller based
speech in is not
i i'm going to spray might talking to
for parts
or
oh i one wanna
well starts by explaining the
is spring that you S C and
at a M I and
oh it's do that i use that is
the proposed amc
for each
i
the duration of arrival or based the or
but to to a lower of channel mean type type to montana hundred much a
estimation um
oh is explored
and
the a target directional speech of cells of progress
probability estimation used also explored
and
the pipe parents
oh
the proposed algorithm we use the at mean in and i use summarised to my study
or
are sent to a much channel based the speech enhancement
oh approach is a a cut to as three into these i i've pass
oh pimp or yeah well
yes this
and masking approach you and class are and but china
the the pretend
i
or or other than oh
in he's he's a nineteen is a pretty issue need up over there
oh
by P
that's to or approaching proteins
a plus the in
this part
so
so this is she or
in
compares the two or picks the problem
and other the meant to use and those can signal
oh at at bit proof of peaks from calm uh a a at the bit
or for a as
that takes that in
in the type
to options so that
no that or so as you can also so to obtain a at to and
in contrast
let look and products
but i don't
to noise show can us so that
whole
not have
to a chance of the nose
mean that a collection can thus
oh can pass through
and
my a cancer that generates
but you know as good at the processing and that's it's probably a and it be a map of signals
so
yeah
use uh
oh you do one
and and don't was set
but the lead to by reason is
for a two
in the uh yeah and and C
or
the main problem yeah but to a solution you is
or
how do it
to a show together or
well
oh have a to assess about special
um
especially uh
but
and they should have had attention
mission as you do a petition that's gonna come from that scene everybody a
oh to give us performance so
use
a just a C P
is that there is just a touch and others
come to
the
get that processing that it would be have and the S C
i i have to the do it's again a to been to both
so that was assume is cut can this to time
uh
hi but this stone and so that the is a
uh how the addition is
there's is um
oh have that
how that sparse component is for me not be it's clear to be in a not so
how this is for about a
the
yeah yeah
sure is
so that have a nice
how a decision based M C
a
and to that and is the based them to do and
where the a should be can target
and there's a clear
in from us
yeah a separate this is based on the
how much the time the speech energy is included in the in
to have yeah
a time-frequency bin
so
it's separate addition for this
the
in the where the tears yeah it's important to to run
a a no some put additional approach
is user
it to as
that kind of
speech
that's that's is as probabilities
so this so that a
this so that the shows
or or i or proposed to sort additionally am
oh and i to the conventional or and a she or
in i
a the area she
or of a probability a
but
the clutter probability in
improved the you a cancer uh and
i have a the probability is improve that you know
i the can the kinetics
uh
they have a that
for a a point of high and of the probability use the plan to ask a product of
or require a and clear
but known at least
in our study
so
so missed stats
or
you start from to help the is
or
that this thing is
no permutation of components
and
this and is kind of the components so
yeah nice to have but is time
to have to a
you can try
the
i do to action a speech to has i could mean have to that
and
for um
that
oh like or you the ratio that we can obtain the core in a probability so i could be
a
meaning to i don't tuition are
components
at that we we need to estimate the perry so the kind of all components in target directional components so
the estimation process to use of i could do this
the you venture
uh a can have that the
i
uh
a from a uh had a a a for a to have a use that to probably for the reliable
estimate of to
and it to action a speech of or pretty probabilities
uh
this parameter one
look at this power but that uh
use the pin show is have that
direction or
to not how to a directional components
uh
no
do a tune and tied it to show that component
issue
oh
is is use the can show you that's and clear that or a word special a
information
so
hmmm
in prayer
you improved to
and an additional uh a no
by incorporating a
but you it is gonna press
to
T that estimate
oh i could do this is up where this is opposed to the scope was to plano
the million
you
to
estimate the the of us to have an estimate
so that stats
most type
a problem
the to been can extract to this that to go
but channel in from up to
so this is that in problems
storage to a problem
a this that so we can
i right
that pimp and the estimates condition
oh
this to me yes
but it it about the function to
improvements the
the the data base
you see creation use
uh
can be you to this
or
at this occasion
so we can try to the
oh
you can do that to knots compunction to be formers
right could these
do you do this you can and me
can can it
had to H are not you know a so and there's a comparison like this
uh of that kind of direction and that's fashion
and me yes
oh to the for mapping
and can promote
and for the image compression
we you use spread phonetics
so
me
obtain a the hundred to actually
sure
that's component yeah
and i don't need to completion
or like this
so you know to that i to two action and is the can question we
can you
we can obtain the approximate but that approximately the but don't know the kind of direction ah
and of to the now and then by the collection of components like but these
the
that's can is applied to a
yeah wrapper as channel signal signals
so you doing this
these approximately do
which signals maybe
obtained
with the
a close to can a one channels
i could be a
the
the corpus to the can only
uh
but a it's somewhere else to the just to
to to has a function like these
so so if i in it can obtain
or
a post so it an i by averaging the value
a
up to a poor
um
a to a like open or E
C send to per cent you may take to balance space
use is constructed
you know a you might amounts
and speech and energy G uh from we could do
and
actually mixed need of
the purpose in on it
a you prepared the
to additional provisions for the performance comparison of you
no you reduction in every and i and the of china evaluation of speech quality or
for the had a dish ten
that of
to as you can interval was may now to detect key
and
for about a conventional a separate the decisions are we
for a
i pitch
you know
a in the
let ones
paper
yeah
uh proposed the
what this is a message
so
C so is a noisy signal
yeah
it is quite a and it means all the white house
is
okay
but
and
uh uh of the decision
but
a
yeah
proposed
so we show
but
and and in the menu or lead the T
is that have the least still
it just the that runs
yeah
so
no you reduction and propose
evaluation
oh about
the uh and that the noise reduction was used to compare just signal of that just not just gonna of
new the than a discussion
the
in the
yes
or this proposed the
or
approach
or
have a that's sure
and it has
that the performance
that are our results
or was so in that every test in uh
or pulse
or our proposed approach
uh compared the performance
to
then you only had a distant and
and the parents then
conduct of the addition
and
this an issue
the pesq can has
yeah
the proposed that amc a
oh
has
or better performance
or compare with people most than the other
oh yeah
commissioner soap addition am C
so in conclusion in conclusion
a a plus to a to do this the post to train a was
in this can improve the reliability here
oh a lot of at time
to the city P S
and
however i have a the problem
a decision is was to
yeah i in she maybe yeah
and can of up to this show
okay i that further the to use in providing that and see no it has a consumer
so
it that's shown per the performance
compares them using pesq a and a every scenario
uh uh uh uh that
i for for the was this time to
but from better than
that
conventional
a a a a a to see this
you
thank you very much
i and a questions comments
from the audience
yes please
a a is a microphone here
i
yeah i wanted to ask "'em" your own experiments in the car
it was it did some quite good but you played that the and what i would be interested if you
make different experiments with those will then windows can would
in on and you know file be on one P um and things like that i mean
but that but it but that you try to do that at the critical points of work
oh of of your to the menu been dealing with the in situation
a a a a uh i just the can the noise
the
maybe stay or noise
problem
"'cause" to system
so does it mean that the engine was not one to two
yes i think and
that kind of sees
oh
is
i you the from that of conventional we know
two
i see one
and you
so and how many microphones yeah yeah for this application
well
and and microphones that we have there
how many might microphones
the
do you okay yeah i'll to for the simulations right yes
okay
more comments
quest
so do you expect this in a performance if you have and the noise uh a non stationary noise so
is like the windows or man in exact or
you expect a similar performance then
one
but uh what kind of a do you expect if you have a non-stationary noise or is like you are
the windows close the windows
i i engine noise exactly
two
oh
i se and do you expect a and that performance or do you a a change algorithm if you have
a
and noise that it it is just had a yes
i i i expect this similar performance
or or or all kinds of noise
okay
okay