i build your presentation about rope and direction estimation method using component pressure and energy gradients and this work has been a with use that we a cue from although universe the also okay well here is to outline my presentation first uh short introduction to this topic and then some background about the direction estimation mean G chi i don't analysis and also uh i will present the microphone a right which probably used mike the that's signals for for this and now alice and then uh i will present the this big method for direction estimation this this come from the rich pressure and energy gradients and also the microphone error rate which is optimized for this method and uh then some evaluations and one of the summer of this presentation ah well T estimation of direction well it S of or or per pulse in several applications like a a source local station and beamforming and uh also in in uh this got of parametric spatial audio coding methods and that there's a huge or a large scale of you for that's this estimate direction like in music and it's breed it's cetera there are but here we are concentrating do the direct uh these sound in this the based may that's so we are using that's for for direction estimation and that this kind of approach has been used with the directional audio coding which is sir technique for recording and a repair routing spatial sound and a whole here in this figure you can see one the application teleconferencing where we have a some remote location there are some some twelve or send microphone array which kept to the sound and and then we do some some encoding and decoding and that then we should have a somehow spatialised telecon for from the other end "'kay" uh so um this uh noted noted and i'll analysis is based on the sound in those vectors so which uh it which are uh represent the direction and magnitude of the that's flow of sound energy and uh this uh vectors are are computed as a pressure at times particle well velocity in one point of sound field and uh oh the direction of the rival it's of obtain it a simply bleep taking an ops of a side opposite direction of the so sound to the vector and um you know or applications a related to do do you arc we have used to be format microphone signals in this analysis so this signals consist of of one omnidirectional signal on and three three die was four for X Y and chit directions so these type they the approximate the the body go well all C D's and uh uh instead of using a for instance sound field microphone or another and the microphones for for be form microphone signals we have you have been used this kind of uh microphone a rate of or four only direct sum microphones which are placed close to one another and up uh this a horizontal be format signals can be derived it from this this kind of error rate and uh the idols or computed just type biting you known taking a breast a gradient from opposing microphones so X type of the wide of are just you one want direct the would each two and and so on and uh well and this W signal this only direct something lights just a and number eight over or microphone signals here but the unfortunately we have some problems with this this kind of error rate when creating those those type goes at high frequencies uh this type was so deformed because of the spatial and that well this this uh the spatial realising frequency here if the depends on the and the distance between opposing microphones and here i have a a well that three different three three figures for three different erase so uh this oh first one here well this is for yeah or with the one centimetre be distance and uh well it it produce quite a it die balls that or or frequencies here but when we increase the distance between microphones and centre will be some problems at high frequencies here and here so these are not bibles anymore and up obviously this has some influence on on a direction estimation so at high frequencies and up uh here is the this a direction well the the estimation error here it's express it does uh root mean square error or here in this figure a a function of frequency so at high frequencies after this or specialising frequencies this uh yeah or is quite a huge and uh and on the other hand a low frequencies also depending on the distance between microphones we have some some estimation error because of the inter in a no of the microphones and uh and basically we can estimate the direction reliable only within a so then frequency window um um so um in this work we are proposing to use um this kind of a array which uh a consist of four four omnidirectional microphones with relatively large housing so so they are shallow so i won't have high frequencies and this this provides us some some in microphone level differences so uh this uh microphones are are run such that there one axis directions are pointing to the a post side directions in in this microphone pairs pairs here and uh well this perhaps just so just a sure rest of how how this uh as sound this shot what and uh at rated because of the chateau one so in this direction directional patterns here and these are for two two different microphones this left one is four eight K G microphone which is larger done this another one this grass microphone here but anyway we can see that these are not on directional direction anymore at high frequencies so and uh so uh this this effect this you the lies here here with direction estimation then and up so um um for for estimating direction we are proposing to use uh or or computing the energy gradients between those microphones are high frequency so it's up just computing the the subtraction between power spectrum of the microphones as that we are we are approximating sound in directly with this with this up action subtraction here and up it produce that's this kind of type will direct direct is for for a for this approximate a approximated in this to the vectors here and uh we are using directly D these for direction estimation at high frequencies and up well but on the other hand we don't have any any uh major or or we don't have a prominent inter michael level difference is that low frequency so there we use use just very shall make that for for computing first pressure gradient and then then in the C the vectors from them so this is somehow combination between impression that in you gradients uh okay well i uh then another i don't topic in his presentation was to you uh optimize microphone a rate for this it computation so the idea here is to knots this a spatial i freak ones with the frequency limit for using the energy gradient and uh so as i mentioned this this is a i frequency it's depends on the inter microphone distances and uh frequency lee for for in into gradients it's depends when the dive faq "'em" size of the microphone and uh here this um a a we no effect four omnidirectional microphone it's uh speech described with the directivity index which is a ray sure between uh on axis energy and a total so energy we just integrated over all directions of this you all some some idea about this direct sum no use of the omnidirectional microphone that high frequencies and uh on the other hand this uh a direct to be index it's depends on the ratio of uh between a die fry came circle for Ms and wavelength well this K A factor it's three that's this ratio and uh and uh after and would this and with this direct T V index and a K a factor we get this kind of this kind of gore for omnidirectional directional microphone so this this represent this uh directive the index as a function of K A and uh finally we can compute the optimist distance between microphones with this formal here so basically we just a defined that how much we want this up to use this data shadow a we affect here uh so we just choose one some that are direct to be index value here and then it we take the corresponding K of well oh here and then compute the distance okay well are then some evaluations uh this were conducted in and a and i a chamber on the measurements were done in and the chamber and that using a a K G microphones for a gauge you microphones with i for i come of two point one centimetres and that this results in a spacing of three point three centimetres for for this error rate and that also using grass microphone error right which has a more small die for kim size then this a K G microphone and uh again we have a this uh estimation error expressed as a root mean square or here so um this results this solid line this is for for using this uh the additional method using those rest gradient only and that well well as you can see that at high frequencies to zero or is quite it's very significant after just the spatial lies and frequency but but this energy gradient they produce the it's produce very nice nice estimation for us and uh and using a combination of these different radiance we get somehow uh reliable estimation for all for for entire what audio frequency range here and the same with this grass microphone array right you uh so um yeah the summary of my my presentation so uh so the basic idea was to to improve T and now this is which is the direction estimation from the from using this a square microphone error and uh see improvement has actually it by using a using the shot of the microphones and this make method and that and also it was shown that this optimized microphone from rate it's works with this spec method okay well thank you i Q and have time the question i i i think about a work the way which and right i were have you ever cut the experiments i mean pressed of reverberation you should the experiments results any quick humour no but oh right do you have any uh yeah experiments uh a and the experiments in in way easy or or really environments oh yeah yeah yeah i have i have a i have a yeah tried this with this a a teleconferencing application and uh you well it works nice this in well i i i'm this in our experiment of that have we have used it in or more room and also some much environments so yeah i and more questions okay thank you very you