0:00:15you
0:00:16and
0:00:16right
0:00:17my name is
0:00:19though
0:00:20so more people
0:00:21or you know
0:00:22action
0:00:24a a can go you know that your account for a colour version of
0:00:27yeah
0:00:28more work competition
0:00:32i'm i'm so through a uh
0:00:34a yeah
0:00:35it
0:00:37i enough to
0:00:38it'll "'cause" using a that's got model
0:00:40no you know it
0:00:43are first and to reduce as the acoustic echo or then i will present
0:00:47they're model be used for acoustic bass
0:00:49and then using this model out was on the on know to stick
0:00:53it can can all that we proposed in this paper
0:00:56and then uh was on the test results on the uh i would and my presentation for compression
0:01:03so
0:01:04yeah
0:01:05it will
0:01:06the are core problem i as one that far and speech your signal is given by the microphone
0:01:12we have a a a cool white white and you
0:01:14and then sent send back to the a for in speech your and this was did you ain't in that's
0:01:18what we do and to we so
0:01:20someone
0:01:21but to mission in the network
0:01:23to solve this for and approach used use the acoustic know
0:01:27acoustic echo can
0:01:28that try to estimate it was not once press
0:01:31in the microphone signal and the where and what the more while is close to white had is close to
0:01:36what
0:01:36we will this the if effect off for a cool
0:01:40two
0:01:41estimate this it would i i C need to
0:01:44and estimate of five the acoustic channel and uh
0:01:48the most common
0:01:49approach to use
0:01:51is data a you know or environment J where we was that it was channel
0:01:55the it energy or is a convolution between the
0:01:57far and signal
0:01:59and they
0:02:00acoustic of of you work with than that of
0:02:02of the down the not speak or the acoustic channel
0:02:05and the microphone a
0:02:08but uh are to the a some by phone arm reach a right on that we use a load device
0:02:14a devices so would be non you know is
0:02:17some
0:02:17devices and to be some non not is that the all speaker
0:02:21and uh these uh are non no just mentioned on told nonlinear environments
0:02:25and here we can see where some simulation
0:02:27well i and would be so
0:02:29in that all speech your
0:02:31some known note
0:02:32a pro three non know
0:02:34you are it's so the output or the speech or that we suppose that they those be drawn to be
0:02:39some known not
0:02:40and we can see how
0:02:41when D
0:02:42none no it's if it in how they
0:02:45different you know a a good soon
0:02:46performance trees
0:02:48um sort is for and you know
0:02:50we use a a a nonlinear adaptive filter
0:02:53the not not not that if you choice is a system
0:02:55that try to estimate did you know what on and non you know cost
0:02:59of the
0:03:01acoustic channel
0:03:02and uh i
0:03:03try to estimate the it they only or equal on suppress it from day
0:03:08it was signal
0:03:10from the microphone signal
0:03:11but this problem
0:03:13this a
0:03:14more to and thirty some form a as we can see your we use many adaptive filter
0:03:19this will introduce some slow convergence
0:03:21and also as and they do not case
0:03:23the our system depend on them on that we use
0:03:26if well we use a a wrong model or in the in on you know
0:03:29in our on your is
0:03:31we not to improve the performance so we need first
0:03:33a a
0:03:34mortar are often only you is in this case me
0:03:37use a model of the house be john we for that
0:03:40a of the known not just come from that speaker
0:03:44in our approach we use a a a a a of an unknown to
0:03:48this is a
0:03:49this come from the does that you have also
0:03:52from some well for no now speak your on a be have seen that this
0:03:56more that is uh
0:03:58is are interest interesting to model or the not speech or no it's if it
0:04:02and we and also assume that uh in this case
0:04:05they now speak your
0:04:07features
0:04:08um
0:04:08so a very small compared to the acoustic channel your we model
0:04:13and you know stick channel which is a
0:04:15and acoustic uh the acoustic channel or and the microphone off
0:04:19so we can suppose that uh
0:04:22as the acoustic channel you we can suppose that the are stick channel is a you know a filter on
0:04:27also the microphone is the you know a two so would it to that's got of the two channel
0:04:31can be
0:04:32more than a city in from don't we far
0:04:34the are you see
0:04:35of the acoustic channel we suppose that that six channel is a hard
0:04:39variable so we can assume that
0:04:41this part is highly variable
0:04:43and this part is a a less variable
0:04:45but uh uh is that
0:04:47and a is a nonlinear
0:04:51in our proposed up what our aim is to
0:04:55to use it that's got a model of a a a at that's got model
0:04:58where we try to first estimate the output of the non-speech you're
0:05:02sure and then we use a you know adaptive filter
0:05:05estimate it you know of of the last minute that's got of the
0:05:08acoustic channel and the my proof
0:05:11where we use this mortar really gets marked improvement because in general
0:05:16they've i'd every you of eigen the channel the whatever
0:05:19of the acoustic path
0:05:20come from the acoustic channel so
0:05:22that's mean we
0:05:24that's me we only need to wind up a steep of change only need to re that the A C
0:05:30and we don't need to
0:05:31we i of the people so
0:05:35you are so we make a compression between the
0:05:39i a model of are the power power of to a power ch on they that's got is more than
0:05:44that to be
0:05:45was true
0:05:46yeah we can see that there are much much but if you and that's mean
0:05:49the
0:05:49first so channel here
0:05:51is that you and to the convolution that unit it chan
0:05:54and difference channel in the mike of in the proposed so
0:05:58but uh are what we can expect that this just one we use
0:06:03a got a model that we have a high or
0:06:06means is a higher minimum error or to day
0:06:10but model because in the power model
0:06:12as we
0:06:13we estimate only one channel
0:06:15we have a there are about to show that
0:06:17we can have also a minimum error here your and you know that your
0:06:21a are in the proposed still so
0:06:23the the one minimum or you higher compression
0:06:26power power to
0:06:27probably case
0:06:30and uh
0:06:31where what's something that we can uh
0:06:34we
0:06:35something that's
0:06:36we can see that
0:06:37if the H change here
0:06:39we need to race
0:06:40here estimates
0:06:41or different job here and in this case we just need to estimate
0:06:45only did you know a lot of the feature
0:06:49or we will go to how we estimate our model
0:06:52we just we first given even the different expression of well
0:06:55oh signal
0:06:56you are white P
0:06:58is supposed to be
0:06:59is the
0:07:00what is the output of the people sister
0:07:03and and white N is the
0:07:05why and
0:07:06is that
0:07:07it was no
0:07:08and the estimate it also use the same we have
0:07:10white we to with put of the proposed on the sum of
0:07:14they white hats
0:07:15we divide the
0:07:17what we for the H
0:07:19we give our they estimated
0:07:21it was your
0:07:22oh and that or is the difference between the estimates
0:07:25the different in that it was not the estimate of
0:07:28a and it is method it was you know
0:07:31you are we use a image for estimation to see how
0:07:35all of it over here yeah
0:07:36in a me
0:07:37in a
0:07:38test
0:07:39yeah know we estimate our the output
0:07:42the out of here
0:07:43can be
0:07:44estimated it using the mean square error
0:07:47by day
0:07:48cross correlation between
0:07:49the
0:07:50why
0:07:51why i which supposed to be day
0:07:54it was not and they're
0:07:55white white you which is that out speech you're out
0:07:58but in a red we don't have a just to this
0:08:01white P signal
0:08:02you G is the output of a a speaker
0:08:06so this you ask the they are that
0:08:08if i day
0:08:09why
0:08:10the output of the that speaker or is high it
0:08:13we will decrease the common knows rate of for
0:08:15oh you know a feature
0:08:18and then we go to the estimated of the people still so to and this is
0:08:22what one call
0:08:23so for choices a
0:08:25for don't in the people so
0:08:27so we can see that
0:08:28it jar
0:08:29so for
0:08:31a it's of the joke can be estimated
0:08:33a a like
0:08:34cross correlation
0:08:35and the what to correlation and and the adverse or to coalition
0:08:39of
0:08:39the output
0:08:40for each people so
0:08:42this a
0:08:43we the cross correlation between
0:08:45the output of one
0:08:47of the
0:08:49P you have uh
0:08:50so for job and the output of the
0:08:53or or or something to
0:08:54and uh that's um
0:08:56that's why an order be but one
0:08:58a power that system is used
0:09:00people proposed to
0:09:01and not to go in addition
0:09:02this
0:09:03if it
0:09:03because one day
0:09:04system of the one i
0:09:06this uh a really put to zero
0:09:10so
0:09:11after that i we're
0:09:12sure how we added C estimate
0:09:15the different should to this is the
0:09:17and the normal and uh
0:09:19the need
0:09:20you know me
0:09:22this to mean score approach we use your
0:09:24we can see that uh if
0:09:25it's uh and M S words
0:09:27on that we can see that in each case yeah we used
0:09:30the estimated of of the wood of or was so
0:09:33on and also for the
0:09:34people to so for some for joe
0:09:36we use the estimates
0:09:37of the know
0:09:38uh use
0:09:39as we use the
0:09:41estimate of the you noise that's mean we need to
0:09:44ooh some global step size in the put
0:09:46so estimation
0:09:47two and of all you for what these that we just
0:09:50and and also as most of a
0:09:53a a porsche
0:09:54people in in said i the or
0:09:56filter or two zero you in this case in that's good of course
0:09:59we cannot not this anything goes
0:10:00we we not change a remote uh
0:10:03change
0:10:04change the effect of our of it so we need to use it as one of the tapes
0:10:07in the proposed still
0:10:09you equal to one
0:10:11now would go to
0:10:13oh test result
0:10:14well we first give day
0:10:16oh to sit up know we use
0:10:18five people so that's mean the P
0:10:21sure is equal to
0:10:22five
0:10:22and and for each people cells
0:10:25so it for yourself a job we use
0:10:27one hundred taps
0:10:29and and for you know job
0:10:31we use two hundred
0:10:32and in a lot of to be used three
0:10:34hundred taps
0:10:36we use a so
0:10:37signal to noise or racial
0:10:39thirteen db V on fourteen db that is a
0:10:42it was to to denoise noise or which
0:10:46so the first one this
0:10:47when we and
0:10:48the first was your
0:10:51so the different case where a some suppose that
0:10:54we
0:10:54how some they echo path changes
0:10:56so we can see that for different point of the it of changes
0:11:00the different should do we had different
0:11:02know the power of it to is the what's
0:11:04we is the power model
0:11:06and the proposed it
0:11:07yep was an more your
0:11:09use a three
0:11:10taps
0:11:11in the people so on five taps and the
0:11:14people so and uh i think we one is the normal one a a mesa
0:11:18and then the ms so i
0:11:20one we went you in the change parts
0:11:23we can see that when the first echo path change all eyes
0:11:26the and lms on
0:11:28on the power model
0:11:29the part and model
0:11:31do you better result convert
0:11:32the propose it model this is due to the fact that
0:11:35as the first change
0:11:36at the first change
0:11:38pop and model people still a tough
0:11:40a good uh
0:11:42i and i curse estimate of the people still
0:11:44so feature
0:11:46and a then we can see that at this point to a proposed approach
0:11:50you better results and to the second a couple of change
0:11:53we you have a terms that for that was model
0:11:56about to do or more
0:11:59the a second uh this so is when we change
0:12:03we change D
0:12:04it group but they
0:12:06delay in the echo path
0:12:07that's mean we
0:12:08suppose that the
0:12:10we and to be some face is in the a couple of
0:12:13and as for the previous case we can see that
0:12:16when first first date changes in
0:12:18we have better performance for a power of two
0:12:21and the and ms that to a much for sir
0:12:24then they propose model try to calm
0:12:27try to we i that and we have better a performance
0:12:30and
0:12:31when we go to the next
0:12:33a cool
0:12:33do date change changes we
0:12:35the
0:12:35what wasn't model
0:12:37have a better convergence than the all of them but the power tool and the
0:12:42you know and and ms
0:12:44and
0:12:46the and or or to go to
0:12:47conclusion
0:12:48uh we for sure and that's got up was to D in the acoustic
0:12:53the cool constellation we have shown that this uh
0:12:56more than is more robust
0:12:58for the
0:12:58a group of changes and need to get change
0:13:01a if a C a to soup of this work is to reduce the complexity of the system a small
0:13:06so
0:13:07one you know system are really complex
0:13:09on also to have a a bit top one for one to people so
0:13:12because a
0:13:13when did echo path change
0:13:15we should do we used the people still
0:13:18adaptation and the uh
0:13:19it's a improve day
0:13:21it's wouldn't programs
0:13:24and you for attention
0:13:32i you we can takes
0:13:33question
0:13:34a
0:13:50that that's good was like a you
0:13:53yeah it's can
0:13:54it's can i
0:13:55generate can give you as a us on
0:13:57local or local minima
0:13:59but so what we assume your that's
0:14:01the acoustic but
0:14:03is not
0:14:04a it's up
0:14:04but or not that's that's on the elements
0:14:06so we don't have very dot
0:14:09don't face
0:14:10this kind of problem
0:14:11or they are what they local alone
0:14:14but what you know as one we was that the
0:14:17it group of these are we likely to be a
0:14:20step
0:14:31B
0:14:32they they what we model
0:14:35as this uh they do not part of the echo path change
0:14:39if we was that this case
0:14:40so we not to have a a the problem
0:14:44a for a local you one one
0:14:47a
0:14:59yeah it's closer this
0:15:01a a a a whole to proposed so
0:15:12take
0:15:16yeah yeah yeah
0:15:18get it
0:15:19it's pretty however
0:15:20yeah
0:15:22yeah how we initialize ours
0:15:24a but uh as i say we need to
0:15:26was on or step size
0:15:28in the eight people so
0:15:30to avoid the uh
0:15:31for
0:15:33that
0:15:34as we need to use the estimate of the you know to
0:15:37in the estimation
0:15:38so
0:15:45our project is set i to see what we need to
0:15:48what one of the
0:15:49one of it that to
0:15:50equal to one
0:15:52and
0:15:55and yeah
0:16:15they you
0:16:16a group of can be more
0:16:26i to what can be more
0:16:28but
0:16:35H one had and yeah it's a yeah it's can be water
0:16:40there
0:16:41yeah that's what's up in this case that's
0:16:43if i H one had is not
0:16:46we equal to be H
0:16:48the
0:16:48who wish your
0:16:50it's will be in the a you know a lot of H
0:16:54and the and verse with a or of that that's what
0:16:58it's not
0:17:10yeah it's
0:17:11it's a very different close
0:17:13if for suppose that it's hot
0:17:15i H one hot
0:17:17and H two hot
0:17:18a a close to each other if a if numbers
0:17:21you
0:17:21this of a job you close to
0:17:24a yeah
0:17:26yeah sorts a good don't
0:17:28i do make that this but
0:17:29it will not
0:17:30i
0:17:31Q
0:17:35but to to you
0:17:37can is in low so
0:17:39they reported
0:17:40yeah
0:17:43constraints
0:17:44H one
0:17:47any other as two
0:17:51yeah just
0:17:52one small question myself yeah
0:17:53i i is to press
0:17:54and and like a three it and it is that is using the cell phone in it
0:18:01in
0:18:02it's like three
0:18:03yeah yeah
0:18:08uh so that is
0:18:10we use
0:18:10yeah
0:18:11so yeah you
0:18:13uh
0:18:13i as is a uh that's good okay
0:18:16and but the most serious question is
0:18:18and
0:18:19i mean try to understand
0:18:22um well
0:18:23a cool and you what is the magnitude
0:18:26cycle i
0:18:27and the thing to note a nonlinear
0:18:30signal part
0:18:32i i want to let that the data which are processed here was from a real set or whether it
0:18:37was seen the said that
0:18:39if we see any improvement let's say between zero and i think about four db
0:18:43yeah a really
0:18:45and
0:18:45is that's indicative of what we might
0:18:48i
0:18:48yeah in a real
0:18:49uh my about an application
0:18:53for L C for real application this so the our is it is form solution
0:18:58but they also need to read the it did you not a couple of used
0:19:01come from a you know or or a a a a a you this so
0:19:04or not this estimate
0:19:06but uh i in real just what we have seen it's to we use in general are it's some sense
0:19:11signal
0:19:11we can see that there is a
0:19:13reorder out T we have a a a a life it's
0:19:16of you know you know is
0:19:17and i
0:19:19i fifty for it's
0:19:20and to seem to focus on this kind of on this problem
0:19:23okay is that if
0:19:25is are the real test set which you use pos it from a um
0:19:28a um
0:19:29a a stick all
0:19:31i
0:19:32it's a far set it's a cell phone
0:19:39so that's like the speaker