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