ࡱ> F&$ٶa|ջJFIF``fPhotoshop 3.08BIM``8BIM x8BIM8BIM 8BIM 8BIM' 8BIMH/fflff/ff2Z5-8BIMp8BIM8BIM8BIM@@#8BIM8BIMX 210218_MobilityDevCon_template_4 X8BIM8BIM WpTPn@;JFIFHHAdobed            Tp"?   3!1AQa"q2B#$Rb34rC%Scs5&DTdE£t6UeuF'Vfv7GWgw5!1AQaq"2B#R3$brCScs4%&5DTdEU6teuFVfv'7GWgw ?6;QlfGh%mQ;"!8l@%)6\j{ !{"WI8=ɒ?VOWKcc[AɁ! 4TTdcCBjji ➪߯n""vS?Sҹ v @W {7N_RLu]Ui_j+|TдiۨHW _GD#HԂw4|BXH-71 VFBpH]=Ɂ pPnhhK$)j3ZQDj+BQZ @! SnNEqBriHB2!z^¼")Ʌ/aFc2FcO0ZE<"Bף5wb +lv@Q6|FI$p^<ǻY v(uRsP2WoBqFt-0Q-%IMӨDΠ(IiC;qh!5.a 0Nr-C@) )?!.>ZLKZ hnC({o )AvCH<\-YO`)xJ 2M% 0. Lp&'P$!4c4)HvGI$ ٿHs*Փzx1%|5 ֣r}=8&eEn)\jH#̮M$F\ZH{ W&jj+jL̗k@k?kU0l]i-l4lWllblblXAlXAlXAlooNooNooN[ooN[ooN[ooN[m [m [m m m m m4m4m4m4m4m4m m Rm2mEHmSmD}]m]UBm-mAn0n!Ens0ooXTop "p&pC5ph| p p p p pp pT4p@69p+]p ]poz$oo oxNKonXooNooNooNooNooNprprprprprprooNooNooN vj vj vj r< r<,rLrgrOrl@\r?rґssEst$8ltxtutuZvZvmu܈wuCYu!Mtit/6s~9rd%Wr wq-q[pqRp΁pq;.q1rԎrϸs|7st`;Qtu$uq_Iu(uV u|bv vqdVv vj vjr<r<r<twtw7xtw4t`lt9Yt;s<s;rrrpr<r< Wx}Wx}Wx}Wv]Wv]Wv]Gv]Gv]Gv]GxfGxfGxfSxfSxfIuxŎyyȬy<yϸyϸyyyyЇky,yx%x~sSx}Sx}Wx}Wx}Wx}    | |,|{L{g||-@\|??|KA||`|u}}l}p~4~4ZwZwmUGwY~M~0(}`6}B~9|.#%W{w{-zJ[zJ/RzGBzJuz.{1{Ԏ|&ϸ|}l}ٛ;Q~;y~p~_I J gO|bx0dVN  |||}(}(7x}E4} }Y}8;}Q<}B;r|k|#||XA3XA3XA3XA.XA.XA.06.06.Vz^N(!I:}o#R~!~8ܣM-h0606XAXAXAXA` XA` XA` ` ` ` 1m1mKa[ #&>+̗*s\0i-[3b3b3XA3XA3XA3v{v{v{D8D8T/OI! ?E66~qX%9xEuFgUIqXj/EkϸĿϸE( 0; _mY / @/88QQ\mLX rl%q~s?ST36 G]aڸOG@$%9 v#ʁK'Pv{v{ @ @ @ J J,GLFgG3Ha@\aV?fwXl8VYVYZҜZҜmlwLYMwP6Jg~9uH%WwN-o[TRY.L1ȱԎmϸ; ;Q1_Ih: |bUdVf @ @JJJ?P?P7x7j4 EY];<Jg;r]HJJ b_b_b_d_d_e]e-tfcOf~ng9[giQg3h8hE})hoBh8Th{:hQ!ho}(hGM0hgjgpff&d[&d[&b_&b_&b_b_b_b_ ddaddaddadadadaddeicjeܤޥf*t*fWN0fy~.fa}ffffяf4f-WfaeԱet}eg]d1e$ddւddւddaddadda<nZ<nZ<nZ<jɋ<jɋaj'j1jƺju0jkkB)k+k_(lcDlll&TnR#&TnR#Tnmˠ@ma:llo kjZ>jKvi?iPUi&i ¡iѢiPilfjH* uj:kl3i)lRp%m"mdm{!mnLn.nn?\nPAnY<nZ<nZjɋjɋjɋlllPl llyWlRp~llqkpJk2=jܿjɋjɋ+sM+sM+sMq{q{q{pZpZpZ+nt +nt +nt +p' +p' +p' ىq8ىq8ىq8+rP7+rP7+rP7+sM+sM+sM<x<x<x<um<umauk3uj=ujuku0uuu )v)-+v|n(wDw\(wywy&Tx!&Tx!Txxo@xIwwvnuZ>ttrKsUs&s¡s Ѣstp{ft6 uuTv_vj)wDEp%w1xxT!xxLxnxvxAx<x<xumumumwbwblPw[  wCwcvp~v{qvnpJuLuumumyʸyʸyʸyʸyʸyʸ{ll{ll{ll{ll{ll{llyʸyʸyʸ ~~dGYTfL  -JWPt~ګ~Xw}n1|KB|~ |A|>J|@|~>| a8}o~~~q~Oq~Opo~~Ut~/V~ oO}}c}fC~ ~.~TK~J=~/O^׽H ZdmBz/n;q~Oq~O ]W]W]W'.W'.W'.W'. '. '. ; ; gC!;LXt4{,8Zo[ -Z+w)YXqUEDzy@yWQB0YGG]G]G]G]W]W]W 6G6G6GO9GO9G`3,+nZJ>);rCdc*+&*v8yl; L\u_WcxȾAH_&cO(^B^6G6G<-<-<-<Q<Qaru0)W+%(6tD&T$&T$Tt@4ɨADŽZ>Y#U&¡Ѣ#[2f uX")s p%6]!gLn1#A,<-<-QQQlP rL%p~@qpJQQ '.'.'.'.x'.x'.xxx~\%!' --*)?A),Vw6>t^w6>t^w6>t^w6<w6<w6<ew<ew<ew<w67w67w67w66:w66:w66:ew1)ew1)ew1)w61)w61)w61)w6/w6/w6/cC/cC/cC/cC2e-cC2e-cC2e-ss7 ss7 ss7 cC;cC;cC;cC>t^cC>t^cC>t^w6>t^w6>t^w6>t^hkMBhkMBhkMBw6Bw6Bw6Bw6@w6@w6@hkM@hkM@hkM@hkMBhkMBhkMBcCBcCBcCBewBewBewBew@Dew@Dew@DcC@DcC@DcC@DcCBcCBcCBrUKrUKrUKrhMrhMsLXtXRLϼtwL)uvLhuL CvtK wJxI%xRHxG0zwUFu-E(RtzDKrCkp4C֫mCۮkDNj;PE/hFh GBh1HûhIhyJ%iaKjHvLbkLBmXQMmXQMmmKmmKk7K3OiCJGiiHiGʂjwG zk/FlEn!EpXEqEEsrEDtXF~ZuG_v G>v$Hv J.tsK{rUKrUKw6Ow6Ow6Ow6N'w6N'w6N'hkMN'hkMN'hkMN'hkMOhkMOhkMOkMOkMOk?OjFP1jV P[iP.iP@iQ8PhQ)hDQ}hYdQhRFgRg:S gSNshSh Sbh SbjOSbjOSbj7RjQʸk!QlZP^mjOo<Oo<Ow6Ow6Ow6O g:X+g:X+hZOh[,j;P\m,kY]U m0]p+]r{]Otdh]LuV\VwHS[`xUZxHXxV݋w/UnufTTusS&raSoS mS kSjT6hUg W%g:X+g:X+ iXiXiX)jWjfWMj4VGk>Vk(VDlIUnYU|oUzqU|SsqU[t\VP uVܿvW;vXv/Y0uZwt[[sq[nq[ܖpX[fnY[`l?[k1[.kJZfjuZf1j_YjVYi$YiXiXs]s]s]sC_9sC_9s[_Pt ._jtM_Mtj_tp_t8_`u[?`PEu`uVa8v av#b=v$bv!cud@uv0due7jtk=evs+ez!rnLex*q{dϓqOczqOczpapapw `Xo_X o>^nZ^Em^{l4]k$X^3j^hRi@m_.hrG_h`h1bQ-h ddiSf=bkfkflZe2lZe2jdi6cibEHiQa j.`jg`}}j` k_ыl2_mZ_n`bfn^an^ancٴncٴoqf/pgXsjgVtgQufuv;f Xwe2=xdLxRbx`w/_vd^4v^kMu^.XuM^7t0]>t`]s]s] g:lٙg:lٙhn\Choj;PpykYqam0qؗp+qtr{q׽tdhqYbuVpcAwHSom-xUn'xHlLxjw/iufThsgragtogxmgxkgujhԤhiOg k2Kg:lٙg:lٙ ililil6YjkYjfkYj4jk>jk(jQTlIilnYioiqisqit\j\{uj-vkvlv/munt[otsqoqopXonYol?o\k1okJnjunrj_njVm}i$m%}ililb$XEb$XEb$XEx<Ex<Ex<Ex<Hx<Hx<HhoHhoHhoHx<xx<xx<xx<=x<=x<=hohohox<x<x<x<Bx<Bx<Bb$XBb$XBb$XBb$Xb$Xb$XmMMmMMmMMb$Xb$Xb$Xb$XEb$XEb$XE gq%zgq%zgt gҵ47h3BiDj2jCk,lZmnI+f=oir`tv)wRBfxSx[xBx/6wuivuOQtw3s1,nq`pxީm{+kf 1izhGg{=gq%zgq%z k{k{kyt/klKfmG#nlqoBq=rܼ!;sSatQt VtKtGtKrks]s'brq*qL;yp-CoDnnH@Fm'&!lz*kNkw7k{k{x<x<x<x<x<x<b$Xb$Xb$Xb$Xb$Xb$XiOiOh)ThVX)g_g(Xgogq%gw+h5R1i'k m%zh`ol5pkwrPEs)Jt;<nu)Ruk4v̗twrH4w xVBxx[%x9qx`xw:w9v/u!v?wZbx<x< mmmrrsl`tt)t)9t6 sd)rr؊iUqpMoDnHnYvmtVk lW3l|:lJO )lk^k/|nk|7k{k~Jk.ulYlv lPmmazazazffffffazazazazazazg g g x<x<x<x<x<x<g g g g g g b$X~ub$X~ub$X~ux<~ux<~ux<~ux<Ux<Ux<Ub$XUb$XUb$XUb$X~ub$X~ub$X~uazgazgazgfgfgfgfffazazazazgazgazg EtEt@CZ]Fkimb u$BM--19,gy9.Mw%^_N[^""CoX@h%*EtEt )A)A+JPHx҃hjOM;tۙx-2rz>jOp!mVK}4Ws\ ) g/)A)A EYEY%ZzF|7"QNR u UB2g-#!,L^u.Mw^CD[D"")op|h |EYEY )&)&+jPHxhj4;YV~"ۙ-2Wz>6j4pARs}QWX 2 DŽ/)&)&7979797oN7oN[Q~cXiuv>ysH.\kZ1R1*9UHn#E<#E<UcFDk 2S=;7:RTHBg&o39n,6k)jLW&#2vM!w* }f o}{fP}f/f6}Pf}3f4}f|f^|$f|Kf^|if|@Df!|f{f{fPJ{Řf{Qe{beN{ep{pe:{e {=d{d{,d|d|N|Bdm|ldm)|dm)| d|d|d|mHd/|K3d|,dه| d{~e#{@eMp{e{{Ǡe({3e{ǠfI{f9{\fcD{f2|hf|-bfB|Lmf|o&fZ|f|f0|߅f|f}fb}6tf9@}Hfp}Ze}dQeN}deyQ}dQeKP}Ze"3}Hd}5~dw}md$|d|Vdu|d|d| fi}fi}fi}fi|}fi|}fi|}e|Re|Re|Re|W\e|W\e|W\fi|W\fi|W\fi|W\fi|.fi|.fi|.d||.d||.d||.d||Rd||Rd|d|ןe5|e%z|#e?}e^}euy}[e|e|qeb|e|e|͘e|͘fi}fi}fi}e6|W\e6|W\e6|W\e6|je6|je|e+|e|Ȋe|Ker#|@e]|jeK|@e=;|Ҫe2|ɧe(|e#Y|6e#!|Re#!|Re#!|W\e#!|W\e#!|W\e6|W\e6|W\e6|W\*kkkkÝkÝkÝr Ýr ÝrÜsgñ9sXs sP)sOĴs@syŇswyswyxxx3ņWxK xZ-ĜSxh.xo|xopPxp6šgx]Kx8B{xw"wO3w/JvvcuLu(|ut tF:sTrrrq%rq%rkrkrkrkgkgkgx<gx<gx<gx<x<x<g g g g g g g gg gg gg rg rg rbprbprbprbNÝbNÝbNÝg Ýg Ýg Ýg g g g g g g g g p<*p<*p<*g 3&g 3&g 3&g ѺJg ѺJg ѺJyy{+sz1|c}:}y~g]~cC~cC~cŧj~cŧj~cŧjy?ŧjy?ŧjy?ŧjy?^#y?^#ysyȩyy[Rx|x# zw|wCkw MGw MGkkk'jG.vjG.vjG.vjFuzjFuzjFuzwuzwuzwuzwsj/wsj/wsj/jFsj/jFsj/jFsj/jFqdjFqdjFqdhkqdhkqdhkqdhksj/hksj/gmso@f8sxmfWsesem`spesUdBIsgct]c>tcb~umbvbvΎbw bwLbwbwtbx!bϒxbϒxdOxdOxd2xJudx dwdfwDŽd#wdwo$d vEe\ucefuFf/Ju4f_u'fRu afufu(g1utgvugߧu(hkuzhkuzhkw=hkw=hkw=hkMxhkMxhkMxdxdxdxddGz6ddGz6ddGz6hkMz6hkMz6hkMz6hkM}8`hkM}8`hkM}8`jG.}8`jG.}8`jG.}8`jG.z6jG.z6jG.z6sz6sz6uoz,v;){Cv<|v=|jv9|Av3}iv-*}Ov 7}v }v }w}w}x}|x},mxg|Vx|=x |nx \|,xOzvxszxszxjG.xjG.xjG.xjG.vjG.vjG.v~q,~q,~q,~qQ~qQ~qQyiQyiQyiQyi,yi,yi,8BIM XICC_PROFILE HLinomntrRGB XYZ  1acspMSFTIEC sRGB-HP cprtP3desclwtptbkptrXYZgXYZ,bXYZ@dmndTpdmddvuedLview$lumimeas $tech0 rTRC< gTRC< bTRC< textCopyright (c) 1998 Hewlett-Packard CompanydescsRGB IEC61966-2.1sRGB IEC61966-2.1XYZ QXYZ XYZ o8XYZ bXYZ $descIEC http://www.iec.chIEC http://www.iec.chdesc.IEC 61966-2.1 Default RGB colour space - sRGB.IEC 61966-2.1 Default RGB colour space - sRGBdesc,Reference Viewing Condition in IEC61966-2.1,Reference Viewing Condition in IEC61966-2.1view_. \XYZ L VPWmeassig CRT curv #(-27;@EJOTY^chmrw| %+28>ELRY`gnu| &/8AKT]gqz !-8COZfr~ -;HUcq~ +:IXgw'7HYj{+=Oat 2FZn  % : O d y  ' = T j " 9 Q i  * C \ u & @ Z t .Id %A^z &Ca~1Om&Ed#Cc'Ij4Vx&IlAe@e Ek*Qw;c*R{Gp@j>i  A l !!H!u!!!"'"U"""# #8#f###$$M$|$$% %8%h%%%&'&W&&&''I'z''( (?(q(())8)k))**5*h**++6+i++,,9,n,,- -A-v--..L.../$/Z///050l0011J1112*2c223 3F3334+4e4455M555676r667$7`7788P8899B999:6:t::;-;k;;<' >`>>?!?a??@#@d@@A)AjAAB0BrBBC:C}CDDGDDEEUEEF"FgFFG5G{GHHKHHIIcIIJ7J}JK KSKKL*LrLMMJMMN%NnNOOIOOP'PqPQQPQQR1R|RSS_SSTBTTU(UuUVV\VVWDWWX/X}XYYiYZZVZZ[E[[\5\\]']x]^^l^__a_``W``aOaabIbbcCccd@dde=eef=ffg=ggh?hhiCiijHjjkOkklWlmm`mnnknooxop+ppq:qqrKrss]sttptu(uuv>vvwVwxxnxy*yyzFz{{c{|!||}A}~~b~#G k͂0WGrׇ;iΉ3dʋ0cʍ1fΏ6n֑?zM _ɖ4 uL$h՛BdҞ@iءG&vVǥ8nRĩ7u\ЭD-u`ֲK³8%yhYѹJº;.! zpg_XQKFAǿ=ȼ:ɹ8ʷ6˶5̵5͵6ζ7ϸ9к<Ѿ?DINU\dlvۀ܊ݖޢ)߯6DScs 2F[p(@Xr4Pm8Ww)KmAdobed@X d     u!"1A2# QBa$3Rqb%C&4r 5'S6DTsEF7Gc(UVWdte)8fu*9:HIJXYZghijvwxyzm!1"AQ2aqB#Rb3 $Cr4%ScD&5T6Ed' sFtUeuV7)(GWf8vgwHXhx9IYiy*:JZjz ?/A`l>#`u=$e ߞ>Z'ӦŒր{8zqao6OSc?=ӭ=!6?_ݩIOZ{TPyauzp<[.=z뻒@qaUR#a6}ӭ4r~OOP xC~W߃s5{q޽Wxa߀=uZ8?}/UXWo{:$=:zOVQj[{˯qĂlX>[{WӨcvZ'EQ6}@:O~9x鎵Ȭ>so{ĎIo{?oZQ"\~֪zn>}m{}}:zsߩשz{}l}}{g ~^PO?~O> Saӯcxaa::so}zLn:O09?S[{pީ׿>W}}:zl@G=ztv&#XWC _mi<{ :z.{w7jdQOd=:Fߓ]xI7?{[}ꃯu}ǽ=q*8>G[\ oWc}ԏ^\ {zn?p8uaZ?8u^ד?@>uoOzկ\Hu-=Њun(p tI+#1z`ӡ סm&Va> sp2GVAO%NGPfr@`JL?ߺ:)s ?XMVVO(<uܓJYic"# Oz245F ~}ԎUZo~}s2Qaߺ׺K1t<~O?x}c{xX}} [z] ~޺X=uߺB.I 7qa$u|PGK FP~'rmr-se>׺ Bt?׺CNI?4}uW{ߺ\~׺{{^׺}ߺO< vv)-NAЋf6ѪS(=S-坴M  =uЩWX*͹oǼ /_dcubO3fG1q։{SVHq^:v_kuRk= q<}=ֺu_~}_NCv׭Rw{9hu n?ֺ ?/p:=:^מ~^)+^ֺ΄}?ߺRo~{jH`E{^^ׁtmo>UG~}jy?g\Z7uæʊ{nmtZzG?_{OvZH+U?1_-tU<,boͯ}o{^EI1`ONv_6q`u##q~ǟ{Z:TPԳ M}hNrha OƽkWL'aNGӭTp[??תzN׃?{CGOߴ& Hן}^NHA7f?޿ߩ׳McNa7?SӤn-Ⱦ|~_?SEuREe%c<ґީ윖\Mg7SHUknJ?:޾ 9,YR?=oP??>Nze/u[O׫ $Ͻ_@}p#_u=q_Huq+uzu^{<>GL_oVXu#ӯu_GVպG&O?Jup?؏_z#ՁR[K}ar?[W+atyeVDNA u px<:+7I`J.u#zjTS‘,jmun"_m[:$ TzdBƒs>NZ>I/,{zhJdM#T: kGGV?]i_m״pTf>ց%S"HOu~%;O'?}Ԏԥ@ [uW]o\Ԏ׿M =׫&aƀ??^ի$ܛ{:\},Nߋ޺Ni0q=th6&,N>!,M]?u+cE1Սt2z:m?@?׺@VT,I~g&tuOu^׺{u~/9}{^UvCqլ42U*\jc{Kc6ƮMݚܒo~ fHO(}H*}>׺ѩrڽ,/n>> ޞHZc5^^i~LRzqzʜkϿu [U]yOvRk>ֺyKmoݸ߽ҽWNq__vc݀D$-}{>]W/o^*)A qӤV=G{SFS p`:@ SDbs`~{ n>N$9Zk{=NSֿ׽]<̬utE&ߟ{:GYfȾo_j:l N^)bA<_~8BTpM޿O{S*?z7cXqbo~^S b-+urֽQ~u${)qk_ߩ׫zIn 6 =soq#ϽS˭erZ搛^߈׫#r 'OǽSc?uzGS˯g%?[ވڏXST|>}c)F?{B=փZz_u#*׾α2__i %2{Zrnn9޽:o\,^U5Ld#BzK\g^dg$d~B.3d]J?ݓj$(F#@-eQn>[>Σ`~aӭ7Q Ot#p='R Ϣ.8}ԊuzPDL~VRc{)t}3]GB I[)#`o_utܬ:49t#dtUOGz=8]iߏWY,qckuz){?ۏ޺Qے?>G^q-vz=qu_͍5}G\ZG_=Ќq~}ׯuSsߏ>E:^PM}^:c _߸u*?}^ðvȬ{o\}t.UNFI6@O{^:|?J}s6^8.v8{{^V}}t#WߺQu~bv[5{׺uCeJmܛMn.m{3EK OJnfĒIߺA6ޭ'(dҜ徼_ߺ@vBZ f6|}uҩǁg_,$q1GXO\#P}//6`:l铕?By{:-6E??\z!I?֫B?>rz%eO?Jש1[Z|P?}:90Y\܋?2Z_{։뵏SoϻK-qwZ$u:(/Zwtak{T ߑo:N0x|}ۭW/{UCO)uuHA_:n*)q׽Ӫ9YB_ߛ݀ ?؋[:I m??o^Y%HI3_}/t=zsP O׭tIRuZ=k];zGQ}ǽӯTt^[OZ0}kWN08ǟ{^Ԗ8_A׵uFmo?oWJ,UrE$oŽ9'T^ftvj2n8DJ<uCYU=dTT;;ʼncnx޴'SW yx&oC׺ou_ǽu]q?Zz:A{o~=莽ꞽl7]iNN2WduGX7ˬEoŽ}w>܏?R`Ӯ~??=Ў6Q}ou[:w *ܓŇX͟ɕ0ȱz[FA~=莶BF/i&FvX[)Ձ}GM-40/tCOulyu8}o~F.I=2VRZEߓ^%r4_̲@,=׭}1IjڡQ౱`=Ԏ(zaR)#]R<59ڭKN%$Y۞i&Hߛ?.I{^R(-~=z׮;~}Ўï.߂~X{:=qs?[ގ:Eh?8鎶#~T[>GV?_ǺqߟׄN܁aOSu%:X?׺@^6xyC_>3-qo~zO ؗM2^G^SE(,9^&bb /?~~y:<ߏߺH4soߺP%o{^~q6Q{~׺ׁ{+el:UV=jfmcAIu+I۴+MJ‘&6ɷu]ٻRѤ@Pmq^FfbnO׺{ӧVS,*@ [PudץO*Kon}=cUd[MzCoԧXkBAEp#=Mz .-'nޝkX,???_{uRkԘE>:N¶?b}ۇT'R Z}kʠ}l?{^:̖?݀U'EA?_U\Q7ß~D9T}8xV{u2;,>?ASק:xOvUO%?{:I7.[} o:nl?`uZ-{:<,mt^Nj\q^Sf0&o?`Z~z='Z}uf}ZSUu}?{T@_~]=Pְ=tEITͿOSRVki-zG^\_Ͽ:Nķ}SSEj2΍,/́~^JJ 7ůdylNjNeWyg%-rIܛ:&=z72ߏ[|׺P}?}lވz㤟_zՁG#ӯWmLӯT?ի=>N.>G[?ՃuհzWHpuzvM::)6w (}֘unu"[Y>A`Ӥ~rn5WzzLk䔤jSyHiߺLKv3?$RJ.l?=xI|]b-P9p>{\דv1F_rn=%<,OR:Z '^6V\uoCu6z?>Ǭ?)^ï_nZzu>rXEu#[#ɫ->CQ:厲{l]lm~~~u~͏׺׏^/?t={ q{KQ$>ݱ0֢EbHy>k&Ҧ?=I׺M׺7_͖x4hhPT#KOk0#^p0HJ*Anx|"\A#'~2N؞O^7u{Ԧb2OOoyNφzy$Ĩ ݿIX*ԛvuFѝog^O \?{^z~~vUR`dhzTu,Sǧ׍\ckvz=g2ϻ˭u5" '=ZWvp ~?ӏwԨ??6zu^ዑ6`W^I}OȽ/^ϧXޅSG?U$ vy!8ϭOZݴS}H꺺cj'6vſ{֞^ܖ-f ot*IǽPtSux={ 2R\qh:SjB' ~I_ިzLu*GީML_Vu+O^2pGߩǮjoS=获R+GMI4,lo:H{XI+:I̶<?Ճt"c:oc |oT#[ވ=,R1GOT}ԃ뉊m~?N3&~R(A,ÀOczx=$3YW{ ?_TE>]znRMM@^Ѥ<G{.~'ID$Hyy<}ЎHXcܘy uפƶTH?AQߺY#ǢzSbǏ׭\r??HI8O?Ot"nG[iի6׺1##?slcϬZ/~Gz/R+ׁxپk!N?o+p&QŇok:Ҩ׺3@??^ë#e~y'}[O⟟u#H??GJauI!CyE׫я(@!_zMy+a`Nkz=o/YKrM^ W-ɱ[t+y߿unO{ߺPɹu7?׺T ~s{buT)S!㟧u|vOᄢ@ ~tU#*'na{iV,<~a׽u~զVt7E:ſJ\~nG TI^IU7־}=ֺTPUJxSp.Ot⮌) JKǻװX M+ GuCZħ~@vhuN61M[}tre6 Z8렼?owBz?M_`=:S@ǻ+Se_v^8'onzW7}{DZx~>==9oǻ׉ @?#ޝMA}9|=ONN6ֺn>jS LWs`u*腋om>j$͗czz֡CS?BOZY@[Uk??>zONQ0{X`9'?{+БhInPA#WtPQY+<ß#!X\A[PgGǿi={W'W?>CֺA}6caB ?A6[={9 -޺OA6[ob_Ao~=n=U7K[?-SӭH?ßZu]d]׋{֞^kpPgYA׺Az u ßu}KV-43fDuz_z;]JJM3AR<= ЯUe$c<=ևץ|4i㢦Pگ]Zv?R[s_u^:_uǮ#iӀצ̅5"D@>Ou#dpF=$څLGwz^ zJyKެhӁ~/~ZzI-*ċq0X^yꫥ2?N?1@V+1ܒ.}׫ )i#>o^=Ԏ :ΫTN3Gw7+~c"O^ˬ${\2OGSWK~۟G[:㠁N*:EdSЅA??=ӭNTr}GވLJ]0iz=``oacB:s+[[ì- qr>zmoW7'>z}=pz}获muuވ^lظckRb,~޶:jq>=׫fDMͅ׹>koq<׺LT=Ϳ^V9lukz^u9~}ueDRUO=t9n٨@LZ^{<0Rʊ@}=t_j\ O}_ߺH$gbI}uߺ^׽u~֨BTQe'_fRc]$P52c{T>?R?_Onָg.s=z=*1=#zv :5FRmk-\Eԕ"mca{ȸp:=?%*z۪֝v#u=fͽ u|m>UvZU'˧(aqo}=9ޝhRl}8{SRŸzjbjL`G^ǭcpW(-uKKW?=ڝks+׺o=kRèp}^f9=q隢s~?xVO{ zICs? ȸz3jM}=ڝhbܭ {o1W1%M@@@kͭy>E,?)ԯZ눎׹}ڇ\ :?ߩ׵u֏_돌sk׵y@?O_o{xZxRq>{^ϤOmCRQTߩ {1Ebr??_z+ïtM35%y{ZjfTJuϨ#~oc._Z)ޯujf~$ 8:eA{uN^WU`t|Y(.-{IZzv9t4T jH_ēki׺!I?Ԏ^I?_zz=`nxnՁ4k#t[Q?H+Ͻquvr@ QIR:GI:Ȓb>pܝGW O!^#f}-t}GVN%^Hbqo?_Ϻ@oy9zcވVX(Ŭ>Aom`⇮ G[8uԞ^F^XYx_B?oW`"~}q u2-upk3& >]oBG&Nuƪ8z9X{lQ]8&^ :ST׮ /ǺSϫ;W=ԏz*z=GoOߟzc*Ouz%}k{<:=p>G[ @յ mm{Z*P5?۟u#tϕ[@#ouz.Tf,׺{) HČ*\^|;<|iƦRNHA`C0}t{*HvP,qMY+9?uɿϿu{^׺u{{^רz?H^?p: ?݀ꅺ|Zz\{OR??Wh~GoǻU'Hy~UB^޽^uݨzRy'݂zߏv0{:=rGa-_lskoǽӭT~}3K-t0ѯ䟯ַ֫ױCW8pvca{~9azzSB TV~6“׽ӯuރk?_ߩ׺T_\~u'ߩtc?Soz߾}zceO$>~^C{_|]o~>mt{%O{֑׺@zxZhg:NxqDmZJ\^yz^^]m%\bDo#Y~Ò?޽获_Ӟ/04$K\WMqz-cP%%58Qm^5=o_NN:,-Žԏ^-[~?:n}֝ZGa{?ӭYR0nGm`9=Ԏr=MdD0?Ï^-[R7F4TxR6R+tSR)f:{f;?ҝ[U~fb&7$TcuzNJ+T97?[[HgYǢOVee8Qt#zH>n4#z{VG[@t#u}ҝXu#ӭ˫.oɻluP8|=чW,4?}׭u o=at j_Hۥ+~^[M]XzQoNì~ouzP}`v_}Xoz\z=paS},ο[5^{*q{5W}@؛#MԎkQoup~}%dۛ.Db?Ooz#t+_=ׯuߺ\IHB9TI'u`S(d' D@Eu+snZj>~-{;)lO'UK{^8X^돿u{{^׺^,$[/??Ш 'Ocw##/Ԗ!GǯZ'BjRHzRihcL Soby:*Ӫmf?ԪiV}utIC1i?_~ΨOY'hXw(sq}CWzцG^ORprs}=Z[uօyg݀:5j?zu _޿{6Mzk{}k }"ӭԔ"}z=LH<{>]P9T([{W\}ڟuZߋ^Ԙ66~= uRMXcW}`:Y}`}mnx?oZ-GM?sm'׬ձ\]o.nm?ts jeH F I{h \ ɄjPb ~/{u@ Ň?Aߺ@C#%K,H&o~ߺLdy7^׺u{{^,0<H]*OǿuVZ89;]P:udlo~Ѩ&I,-uEI?y::)=DV}+kB[,svzNH$n`??q剑{UDk[?Ry<{]g*ȭ`uBkXdh_?=:JuLn_^EE/Cc>^>EF)N`:]\.Iow։Kji0 ݀jzZ%[O݅zRP6aziӄQZms?{˪$ v׭u9׺PZ)I-lv^JTڔ^82?{+I(NO]{:\IC[[tEVUǿSj=("p9~:OY{Tċ}{=uO{^]h|{:O\YH~^ )z\4[ވzT׷>Njm^YF [O>NPzLIY?(S-eDuϤ\1)@`3[Tu#PAWZN$mSPXmN^ipp^i/Gz#?.*(idT0WS'!Soi$tQl\ ׷ЎJK+\[Lk G*][GrBXɩplRoMvJ4G[^6Jnx~kdRX9}R: [lg}{ )l:Q_luBXī_O[SLy?p9GNq {@sq]oŚƮA83TP@ A]9yըjH^8Oɷ&zI^fןzO?#޿׽zMKQ[PSK#iUP[u=olmj=L$2}!ԣbmq&g/ -e ׺^wbZ{^ƽuߺ^׽u~rU,lҿoi_Qmp8#{׺Ҩ}M5<ƿR{pOYK{T"O)>uTߓz͡_b?}}uO\vow|Sc7C<>{pct|y7RӨʡ=h:X},T&f'Kޝh㥝%L5J6~?7ju^7U'Q'J%6|f݂fPNjU'g?>i}zΪGs{1j|=Imvdj[vֱ@/?_<>d ſs`޽ڝ{ SUd6xN3 Mӏߩ*zK#o^zbq׿>Dr8)uꎤA9>N_Jj:ˁvy}z.H?t=^N^$}OZ\lM:zG?_OZ }-{^ˬd7SztYPY>̀NävF)>Hć7ۏz׿>i# 5a6TT'\l>NS}sT[tZ-TkQ  GoAީ^=$kus\I7ک?:?\z=:Kda 72HsG $=֟.Iven~`b۟zV0H+;-RN¥uG}E{P?Oz#׫n-}??^#|׹׿SՁ?B:i{8 z o>G^F{$_R?oV~5ol^uq0˫zu}ЎQ1NupIc׿<{Z;2 x=4׃kg׬J9~GV:ԑׇ͇\tPOg-lzu>ЎìXSzpv厣OǺD?~S ӡi}(,ue4?ot'RInOԛGt{~QucSNlc_cO[2SӣI,> W wom_{(kU[ߞyu[s4$P9. ^o׺ k+V7k6y^顎~c+s{7{,Gן[ߺ\G׺ߺYR&{-?^EΡȷ׺X܌{0׺fAW Plt- ׺өE"+5M')dxzmOY{S*O{u{j&\O> Ei GA\R!sߑuӧiD%m= 5=A[8E=bh4z^J/[t5}nx>TZ[=SgB?ߏ}h!_|}P24?NO<_my"hYPwh!T?ϽҽVCI7~?{kO>^:5?_znn}ڝkm{?>NY#1<N,Muo?ƽ{`+؟?4DxG_~Zzy7?{[ֵt~~y}kWOHߨzqTںVK'+ںSs=o_\Jq~~:ޮ'+:}~_o2U H+*(US^r(cpd,m~}ꇭevbjDfV?vވX-[.6jnlR:zKTj%iumRxީF>LI71,Gok{zz0]qy{zrp^?~JRnGMR޿~}U$?R1X%AxxՁQOoիR}ϺᎰ?GW-_@9{nܟ=ЏOGީ^]2,EV#}Omu`}zTȵ؏yuFe{:pg~ߎ/Ӯ>K\+պ~aۏmǧXXxO{{`BO`@8P-\y68ߋ[a3m{{_f0O_0|=b K}ӟì PGXuQSոߏ{x+HޜpzX`^ 1ICTz<:HQ%=ҝlt]fUុp<=H~cꧩC u*ږ?K~_jdtטRnq߷G)dAb1sԮ:PnC}V[<4맧sa?:8d@y?^>]k>*T#MTtx[*TsԕRm#]J@@|mc`[/U"7ߗU'HPGָ?|}ڝju݀ꅫì=>=s }/ӇY}ڝ{'զOc#,ۛN[OϽӪj9^MO{^Xno{p~mtX;Go~9NQX%1coq(*o{˯Ttf{z&K{cTumuu}~ηNsǽSg=Y-+ H(ީתz 2Y/ΩDb?-ީՃtD&R,k:AƎ oA(o;G[뉑XzW`-qo=֞}{#N b)}莯Px)ΰ2sF[HoM7ϺpxW}R+ÏVs[b+oGV.>#?.#?)NGOz9[R/)պ<_OVzQW#??>@`aqmmNq`h <[&O[˨H@?<_HV<^϶ոԐI&'=ҝluX ua׼??C#:끱~[߽u`z'=ҟ?gץJpTik ŽaukS~Glt{}𿺞:cOcޏ[2-T*ZF`$}O{l&bqzMC3|@qͬy׺KOVĨ$x{CՁW5rIG^X}b?ߺuÎ CO#qֿ\r=t9 o`AmcϿum5{9ߺOP:(G?o~=Sl5`Y}tRBoΑouU ~}uĿoE~yZuթ5kfUߺRTƽTHoۀduZ Wz̢ۀuRzΊ>~_1kԁo? qc[p:Bѷ/<vU9C1 ~~|{pOJ'*b_Ov@ln=o ?[U'.,yo{TRu"_-w_N*oʑc{:=HBIwP GA-uRz_v﷮`ƅTY L}Pݩ־g[תYUHvU-oͅ{:OXY?o`uuH֋SRG_^$>[=U뎐MϽ<`׺{q}{QX%da/otzzzzP['Zo^Yn}{{zX@\_o^޺OL\`n{[!XF??ouytx c+؁Ou .{:nDuz?޽Ўul֝zxG[8uOlՃu>:& '?OuuOz#n2dSg]?{P:@zQLf-W~sp-'rU}B)/Q)Ձ;rl?zz;?_u#qGm+?OSue=Guf>GNuۂCX,^??_t#ӫ׼Jq:p{u=bd$ßm0O޺Q~X?~`%@R+ӀW'7NQ@bnO }=8g=rŗlo^'~#H S[t=lcS C*_~؞=zAVv??Dc3}O,=G[7xwƤ6\xC)tͯySOrrջzώߏu%*f qjTH#'؏Lj~^z=+q%UgJ0[<z׺^PllU"?<ܽzJhqTx@iQoz8+ "m{׺?x^-xͽuהp=u[}uɫ~ߺ^2.[ߺ\֩P?musR~}y։2o):݇U9Bn։YVߓ=T-݀zȧH$cuiuIBuORR5^mr??:ku $։Z#}zth\'k!:֖ }ۀSNm$NMqu?`?o^ԅ`:lYA@8_\2z^EEE{PYuOscT[_{^'_=Kc|}IUsn9N@uHq{꺺P}z/sTu G^SH-_~ X}=tKR8_~{9ߺOY'˯T}׫j|oOǽ:&#9]#WL>:ۏQ{DO4L̪mv<~?ě{:'kѶ)uzbTy:ps>G[ ?~#ީ^&}B:G^{^Ǯ0>_S[ad{R8竌`tǺCFuV:br=E:\L_[{luyOt=ow>uԼlZ?ACSR*؎SC#1fnQI*'CNAĐԎҮ \Ie`^}\ztdd%cu=s`nOS=TU$[}Tc<ƽ֎R?⃟6Mz̼ ?U'}^zΊ'ۃ @@q3{Z'|}a7~Cǥbk.Im:MFhsu`:=M_ֿ[n})zg0uX^?uRkå',*Z8+ӑuӥDN_:lxu"Ӌp1װ:Ȥ֩@?oíY\OTzʫ>]{p?~'cWu{^O]X߁te^dCȸ oKOs#u <oqt^Qo~fZ{aSTSqleyy׫(Dr>{^O4??T;^y{֞>zb`~?GuH\)s]l5} s rGވ^<3* s{R:IhR??_z'~?[?:X6U,~}Ԏ&ɹ_֘{վcE4?/ֽcdWH£ ?Zt5X1'QCa[|>θx9?<ou*߁ѓObO'VEɵ[o%{8 z~}ah ۟:^ zt#F>~ζ2Cպo?\dSG*Ko{:+$}lqChےI}Ou`}zæ>í6?޽9^q?_{׺_}莮:WCո;yB:=a'~=}:J,+ {zn.Ϻt$zGTz="\_{9e}* $ۏï=,XؼXV=7տ>*d{b gbY7>n޺AVB׺-R!ǽyuo؏}{l㭎K\px>zpu@%Tky~={Bmst={yr6=u\n=u ۆ?yc~{O$_uT[?~~{^?W{׍A~tno?^׺~[^ЩUI@qX\uN  ݀OYǟwԤ_:z=GU_w1:/:ǩh,?`:Ϯk{P.cv=Wۀ|R\Xk릈a>|s00W{tDKupxWhTd\I9{:Ntu48|RXHN?q։:uF._<$:+<݀M|T(F(7ǻ׫NJ~?':*?݇U-p|?p:?NOR"JOU'a@"˪YG=WDpxOi=zz>N^?tUpv,$kWQX:5[WЏ{ޞ^"}@i?[Z{o޴Lo?OWu-=^֞}޴:zB~?yu`GX@}o^8s?ӯu[oϺM5k8#O$mZtd>6fUߟupzKM!ooG)ƽo ^QV<*m:u`e6>tquo7\K'uz}I>G^~]FpקuQH u#׫u/#=X'?[Hu~c?¢~>GW%@Zm}Gu}>}lun}G΢[[^2)gPcu=Daӟ>9zzƨH<S?{lՇVdu1=?N?SëuŖ}GQ[][R5s{V9<\l7?+{Oׁ>#=x`E\=ëŠT܎OqXu['^OIiHT$ÓGN`?4W}Il3T4i@Q ~\un|4j+I-Dq }׏[(& 4,q{WjGsO[Z~m9O^)k=[Hl[ȸ_XuT{I_=uU}?~yS׺*6^~g?~u%@Qaɷ>׺, ~=uf|A#;[ߺY"ߒ/uT?\^ѩA߼c=JEU=IQ?_U:^~onuauCƝgA#Z'IQuN9B?wT)'x񯯷%V>NԄ[X]W=:J?~zYGZ'qx|l~ޛ~>JiIA>g=T&@A qSt_=R'Y~_t5]Fj#C[X}^t˒MA/horqoϻְ>ޟ12g!\p_9}ޝPK5޿RQ/uku}/ub~q0y7y?JW@ݴ{fGꥺXovU'L?Ǐ{{jpI?']gUB9Cn>z^GC͹ǽ׺ʑ~׺o{׺O~׺_aߨ:M54So[oz^MG'ߨzNpOp9׏8O~=֝{T|:},`aǟի^6?ɰ {UJ!_[O_t[ ŕ,ʼ o~-a]lthJqΞ>|=`tSo=>}luGWH}>:D^zGt\Ghۑk~~duu56s,aa=Xg ǫ)??ч[2} 9V_ƽu<R1"QXl/oR:?έ}GV+,DoXtP#i$~^GV=xE~YH2u'otV''K~ QucҧC4~i $C垶8u:z/cn7~?=4OTy>ՇM2R?޺PguG[+X_~:;~Ocxזc? OG?ׯtzy<\ȽuGSuI?^G=u<lpn޿^nn?+u{ҨY!?btNMU@:DkoW?}:Rêvɩ2l/s{}}:R@ SԘd!Vk .}:=fwf@m^}߭uG:5R/?ӯKN%Yc׸u']-+{pCӑow4Mz]RΒe`kyp:#x5b=R¾5./{{>Y}c?%}F^AR~+7+^G}WGS^-`+eO{~{(??$R]sG)z^q=ۭuX ݴ0P{ }?z=yI[sosoϿzeqm{ZpS׵u{Z׽5ںO^,}zl r>?Akuӡkco~}cg_9<ՅͿ׷~^N׎ci6#?W[;/uz$g=׏WxtQN)9H$28][oTc&CO^\|rmmdS֙B~aO':k&MGt`ӬatXJuיG ў}Cׁ3 q:3ӀuV?ō=zQ_[{[ ~П׺=`d g׫B:3Ǻ9u?cQ[m?=ë'?g z|Xd?~?#Va_}yzs[mVEG'g4ԡG[dgTd{ ?OW5UP]CӖ.3 _HxStOsͤE^-ͿOÏL?Fhヌ[Gzv's9 '\}@oӓ#uXmN=ӭH_'zpzu94~<_u={fP?o]{1X{^#Ksɷr^/Ql׺O~g <~fC^N?ߺ_Өhk\60BmQn>}OUn^zoP /k~ow1뚯w7\k}ޝWϩ{p uR}:=ԅϯu??_n=6Nz̠p:Rq=f@'<nRI|%#R9Os˦惮 LX7<&ߎ?{\Ϭ9F} }ۀu.Ե *[ZO`:5麻Ю *\QztץLz$P^ =߭Kb4hP?p)IY1Y3g䛂?ݩ zX#, .?޽^[ovíov [Po׻׺acu N}=ۭTuKs^}?_+gZFM_t]co:\Ir/~R~ugDz5xy0E~s?}{ }}M:zlc6[JuM_8so׽ԠA[zs@>iGb__u+:z:=6@\qzIzK(?>˭פh&?s{z5 %,?&GW& B):ʣ{Ϭ,t#Gu}`zR:Ϭo_?t#Nt#:"scl?}ЎXL`'ۭ￧t9umyB:B:W+{ln1{B+{l23v@+GQF}\uE6u?ӃQN=Xq7}Mx NہBluL+p?}zM@P~oHAAaԟtzorұq~I{[}IV{׬/ۃSԄ_qOR}ϷU=IEQ7ԥ?_ۃU?>QǬ>{TR ~U<:Ψۀc"ȽsԄ_Ucԕ_v|Y?__Z.IOM~^:7Y($~n,lUǥ4">X~޽ڝT }̑VQ=㎮I+P*^>ꇬ P醖<9^uc^#?7Ou|VFEG:zzߺz_{^)>C׳܁5t) ~o|Y^zu~ש _Qo:zȏ?G߇^J[7maTuM( ?ׁ鬫FN>Du&9G[ GX>J"Fco?O}Fp ǫ|[DWtRo7M} qXY5}C^W~?~GV@[̄ոרΪ>?'g/qǺՔǶ}C5Ϻlup?Nbqo~.:cuaKo?}zQ3{gӯu9&}Ў$?So}o:cӞU -8u4VE`nob?mNCoo' Ft6}8ëqcsmӌPHYy*tOOGM5 !&POlaTj!Vc϶OظiMz=f6~8>zڞ*Qou#e_OV<7\uB#{}CvO^ïuܛ &_u@wm゙׺XuߺRo^Pqn~~׺\<{^IoH׺թpKoͽǯY#SԄ_۝UovDH=M|{Z9J P(>On(g'?^gZ:/ј۝Pԥۋd*5[wT'IP=>u%@\ԕۀu&dP~I\O{i1ԅ^s:2?іߎ-}?޽۪^L0[cҦP[k\r>~]E!]//kֺ݇qYT+X:sEU'54 oH:Qꞝ> > >ZM YT2XB?֏R=O\:ẕ8Zy(=V]d NC0￧ӭA?##ϽֵuGzu]_>No_o{Z9!o싟A׫:"e?_Ϳo׵uG7_~[V#o}X7R#kqJԁf}ީשY jCתzlhZ߽G[zȎ@ۏϺ}cßt"X?C{` ~)^Y7{]X^R+իM 76p}ӭ}4]Fq"<l`+sǺXZ2_h^4`_˫XAxè?~9NvNt$=SJG Ϻ7NtQ+9,{eoUxOzbsSD4Dq6>z ?Cǭ=Fv?u=82:?[}yzbl?ӫ?{l.:)~_޺˨@_[,a?[}um:->O~}_H=X}^~a/n'u1_~HF׺-b[ߺYPf\N?^֩du.G1G%b=:yۃIV <g5pIӪW ?゚}8u^cP{tuRzǯTR-"[IRPs{{ppH nݔu^9* y10V?O~xc~ޫu X6OpT|^-ɷ79Z/GGOZ?.OvWB8e'}ml6VwtPW>:=G!YT}nS*ѿ*x}ϻ $Oq~zO4o݀T'RyVox{)ֺ:n${:!תY@o]7ȗ [GV馢ǺgCSE;G:s_=׭Cg~B:#GkOgx2ǏCX?g;/B:s:"۟t#Vz}Ձ!P.?~}zQ ~=ЏNˬ%'ǶPO}/mqu+CӀmƽaV^~}z>]`1N^yB:X_duqW_mu~!TI~upzJu2ߺM$ߺ0W5T3zm/.8 l~}ԁ-Om8tYB{u7~u3Ot0Y^Ǩď-Gè/W#{}Ga5{gX ~ߏu#`ZOuX^ }}u?~dCm{(x^׺άa{ˏ'ߺ_ש{P3/?<8=IV}n=ub_OngRs?oT%G?U=JPG}JQI~q6x)ϷZZ=IEsY}=ޞ}Tu%G:oR}=Zu%7#UctK{:n~3zΫ:=J:#7EjLvKak>0sp+ֺ-.拃{[>:iy_>z=egs]FU;/GUz|E}wu%@an-ǽϪdGv^GN^։=r ?n=hsQ}}={\[ߩ^^ɽߺX_+Oz^qϿSTQ:[=ש5o~H쎹ou`F0yy}Ԋ~=@d*-תG\yX9[{U_ϺՁ?{Wz}ԎR-+{:=Et ^c"at}B3ӃLߏt#ϫqEt-{?_t=mOX?^GWQ?폶[ឣ:7>NQgE=:c?C?Otn: Cxo {^=Ѹ3 #:QY7?_z=Fo~G}\{CFGQ\_n4c>a8ǶVzϺպQiԓfs?qg>kݛn{s)Duu4?h869u"׷7c?VCXsw`? NFg{˨ߟպOsѺQߓ88u[?>/t?g^ ӂ.?}~NAoGoz\?/~{ߺ]cp~~e?~wǿuЩ`?y̽bԄOORP_wT)ۀtRƘꧩ(>Rq~]IQϷTZnTkԔ#bwSBGϷ:_փQ6pKXȷ g'~zGORQkǟ}.^=HU炙֏RTXquBz^?7p?gUe9]{ƿG݀PJ7Ӫ>& 86?dxVP[}ޟ`unxzaN'8a}GvP=HT}#nVfT?_t?ZYt]ri떓tjw߽ӭW~^}>O^Z}nGq?>Vnc*ׯ zVAuzu/S`i^2|:MVq϶[GuZc׬ .=ЎmׇQ?N T/[#t$wHz"XO=XbE`?tn㨮?>z;|A4=~}YMztG{zQd}W^omD`A7V8DcuAg8CqV+=]OQ\oV=Caɿ0a`IGyu0H>W:+7'=XtS][>~zv{l|u>zח'}oom:{?omm?o^^0Sӝ`qlu'OVa}pzO_u=oLWyC_}7u]XuO^ߏ{~׺u{h?޽uߺ^ѩT΁.-ԤO߷0:m=::{z\cnIT=KQu\R{飓Ԕ_͹}׸"_OnuF>IQۃJUT똌ZqO>}Pko:R7ǻIBITBOnN''n?{ԘwMK_~MzI֟O9^5='pETFusߺY[U됷T`:O\uqGq`puwaoߩjzNQVױ }9z^=`+T>p+TתG\:zVt['{^`~#թ׈o:*\_t6_LuH<ӫulX=_@[.}^;7__Oduu=B{o_D{VFm asEu#qDu?_m/yS8uoO^aomGQdSomJ:Hn8կ?Ў,l}zWSmi?|zϨM8:"Z`mN=Xzu?:ďCӞc_t<W?؟h#2`/=б&^n:y^yy|}?[OW_N0}Ï]u{hV:ߟ}zحzÓGx7}V>~][mՇQպOt9=o68}zcS׺z\}u~{ߺ^ׯ{A{ҩ.Դ:RnQ}{ppCROw"ꧩ@{Ԥp:k ~.Ԡ?6OYaֺ8֏ϩ ~=RU/:iuZRͿ_v`=Y}:=g??<^!#=\zױEKa|:=d O߿cUO{}:]״}u֏׺S==>X+֝{#e~Ez>]p#}`uXau~=bkVn}zGNqJSfSw۠WGaynm[Q_}"NFqqǯu>/}ӫ2a:WQa8omug#4+OV]E{l΢?}íDomΣ0o^F>z@?ߗ8V+t=9IO׺uafuag=lyO>}Fs.=p+}}ՇN?)Ï7xn:??u=Yz{}Fg{}::?Cg=Fw^&y_z;ƾd׺?}׭ƽS޺suAo^u{{^׺u{{^өtgR{puéj>>]48F=:ԔppOn:?^znN\c}q֏ }}KAv^Դ6ǩ(=5zZJ 9zԸŇ.HA>˪!n u( 2/nu^(WY}?}uRz׬Du ?Z"TuP:ȫ{}vdӬʟOk[\:^떓tj{O׫z}jtW:z=nq*Du?}׬L}ϭ"=Ԏzo"R)o[{[|0}Ϻ3}otFz=Dqc;Ǻu8GgQXs|Ot9?l_?VQ\{o=aW}}G[^=Eq}=Q$?}uaè}8=]xukFl9#GV'XuAF: EoFՇQ}}q^G>r:kikW:C]o=qy?~}׫=Fa}CQ϶۫??t<:3{[Ϭ uu;[3_za?Om8{}{XZ޳Nz>}oӯu~{ߺ^׽u~{ߺ_ԩmU<)Ե{ptdcwI}?UJ}Uu1'ǟTZ=fUy}TbwSʽJQ}:lJۀu>{puF=JAꇆ:_֝T&1MH9]JRX{pu??n}߭HQ6zǻORq֎:zhGqԔ_ϽӪGꧬ==fU}1։2۪vׯu/S@[ߺ~}s}{O~z]}{%?usuN\HR:Xo}S˭oG"=ԏ[?.`:zOV`a=N Cu|HV`qm+u#=o?゙=lu?゙F:sǶ]o}}8:: ǫ)+t#Ny}GyǺ$KպǶW^7ʽCq}Ǻyuu3oëuO?mC+t#>FunV>]Dz3=GW^GGqϟW+OV;{m_muǨ?ȹa}u3CՏQպ}_Σ}aoN3~Xޏ[돺׽u~{ߺ^׽u~թ}bԘǷ@ԴS˩izN=JOۃ]IA}GTcw:RF8puCԴ}㪞JO.=JQ}^h8ǩQGT=KAǻLc֎:=WԐ=8u |= VBwJQ+|]HQ[(Pz!U']fQ=:Yp:J?gT0=OYBTz̫]swovZ떓{OG:]XN=k6ZލzXY_z`ӬD[ވ`{v_m=莬XX{gb#X}u0Gq}{jz| {hqF^_>#Am #Cǭ?pY?pzg=Da9fFgYᄒkF#=\q#}PՅQgW,guWgӣoXp,}z￧:PsGo>>}X$ՇGun8߽qcQǧ:iռ<[^IV+{Qot;{Ǭ??s::}7}GpsG:X:X7XuϮz^׽u~{ߺ^׽u֪_yֽbԤǻǩH9?۠cL^m:nԘΨx->UZ=6z}p::ӎ}>ΪKOߟnԔkR}>?{x֏RsmԵ{JXs7RSB}:R+(}=uCU׆:k}תwϭuVݺ>gQ}vֺF?-juo#uZqU}.^{^ Ouup^;~4=b#k=׭ouo.0B:ot8Ea#o >:zw;QXp}gzQ{lպgoV^7W:[oV^H?~xWzzuCӃQ{h.H=z}muq6tn˨}=ӝDG׺~]Xq+WqQϺ[N?uu=oG?}竎7|=˭O^׺u{{^׺uתg[zᄋzmU>K_uKOoGM)>^={Qæp:R'ۋg=JOnJOnԔcLJR~?=tx.1}z8RԵ}HAJ~U>}In uCԤURP{꧇RTZM!nZ=IOvzY{s~}:oU*=yuRz΃ݽ::U_{j)ּuIjzZ߽ӯu{:^u|=׫_V&Nvz=caquzm3=aǨuǨ>NFqǭu~H?{㨮8_go ÏP΢>}׭gƽx~luWègQmӞ}DoCëuA.]DGgi~l}Do7ND>4Q+׏Q[gQ$?VFm8u@?|}=lzuѺ+N3ÏQ{lQ\|=èӟm=[SN?Q7N7>sQ&lu1}b￧u{b?￯z}}˭^׺u{{^׺uЪ$'ukשi)gRc}==zyu,{qzb}=zo+k{pq]KNZRGu1>8CǩiMR{tuu)-Uu)GۋS#q厩ԘۃU=K_qRXϷUT|Z<:_{pp龥݀OiORP}/ׯRzΣG ynhϩݗۜ::=}k?zpU:[Rz`y^]kZ?゙uG~]xyuHpaXx>FzXyպ}~8xc=ëH?[N:z+ƽu;{l=[ϨSպ x}omW:B=Ѹn϶.?,{o׭mӝDG}>zqz=Xq=E=WmӞ}C?.I?{è{l{t+~O{m[#}}èOWDm8<:pOoLqFm>8IǟQ0zӫO{[c}tQ|;cXh㨯SՇGGot=9w~}=lu}n7{?GooMuX}>6z;}>պuz{ߺ^׽u~{ߺ_Ѫ5cZ}JOn#}nz_n/xu1~׏M/n/Z=K:\OU<:QæSתۃ۩I֏RktuVR{ߗSUL۝PZ}~}:R=xCǩ#?>Wh?sZn'M):RSԅ}aR߻:}::hpJ{z8uG{Mv{?v]f:lE߯uGߟv]d TygWl޽׫ |u0^gX[ӭxuq =ռQޏV{laV}yuǟ=ӇV^7WGQdG{_mu#~z϶ۭO߶QdVtn} :Q\W}=zWcQcocO'?c]\p=n7ۧQM_un: [_.QVz=9:'O>xafi>?רFèNyuKfzqW臨3Xq=xqW狀8c׏Q|}SMպSǫy~uc>zXW^7刺ul~z^׽u~{ߺ^׽uҪ%khu->ۃ}TR}8uFb?I_׭Ը=:oΝKδxu->:oia|Onn'Oní.G:u)=:b}=:R}nuC)>׏v龲/<~p>Yׇ]{Ǭd}:A]Z>{u0#7}\u}h{[9Rz =~]_˨?\:Q#ޏV;yg/QmI}}˫uq\u>oVoQm]o}}zppD#>}o>nC>8uOzuaǨÏQ[ӇרoCվ]E#=DoSQ[[g#lu,?O=\q{u>gϧ|'Ouaf{huaW}c:ߟtn:'tQ߼V+ƿ}:=oϬ .8u;{N?oNF?_z<:XǶ3}}Xt={mX>?t={{^׺u{{^׺Ӫ$cZ}KOnֺhu>׺^:˫uO^}>ή8u}.O[;}ƽq}u78uzEoϺw}xC^Eo￯:Y>պGרCռ>n:'˨ǫQ[DoͿlua,ΣS{u~<3{lǬ-uqG8g CQt=9t=mz_u=luN/QaO[{t={ uqv}hU׽u~{ߺ^׽u~Ԫ$>gZZԤ}=:㎥תK_Cԥ~^=WQ{sG۞TR}:ob{q|AMRGשIۃRۃOwzR?=:m)>>Ե˦RWֺSԄtpK_ǻ u_>Un/:?Wg_wT}:Bȿ:}HhB:cuֳԀocow]kx}mǽZY5֩׉{^1oLn}=П[`c}\z Vas)luSX^Fou=X`o8uOǶQ[OV<:t9fC?V<9Vun=Fcc}}ӝE+߯[EoCƽ97CռC^9Íz'}mxl}ОY=Qǫu?/ۏN7coFר{o׶Ǩ$zgǟQk:Qf?{DsV_Sv}^zOgW#}8u;{.7:~~}[m7ގzXu={mW`o?t={{^׺u{{^׺ժ}XR6J=?U?˩u꧇RSJ=Wqۃ˪G}T-~ۋ}LOǻ=W\~Ԕ?On^:hZ=7Ը)Ǫ*3ԴꧩK=P%M}8^sԕ6>SH^:J{Z=ISM4J=uǩ=~}T}uQt[}:@{=k~[ISqu./=߯uOGu: qS@={Z>׺o~z\IW׺׺\1߼X GboK{ՇXW;SzOՏXOVGsc{VFou=l>OW^cQëCW:ǃ"Fo۫=EsϺ_|GV~}F{l_oOV+mt=XuV=mzgϧY}:'gmӀTuޢ7gIzGlaǨ}CӇ˨?޾{l.VQ_Q}\q4_^g{EoSǧGQ_gcX_պzv:;~_t=luO&}f:Xuzn}>}[~z^׽u~{ߺ^׽u֪c:XԴ{Ԥ[ۣL_ǻ/xu->{tq龤ۃ^TWM<:wk˩KZR?JOn/TR{GR~>Ga}zGTn݆zRԄ?ϷGTLgnԑgH:J=u.O^ >'Nz;mu>9}^G'WV,:tp3?GoϟVDoCë?゚ue45~}sG?mpp+m8ug/Q_g ÇV,tI?}z#}狀=9f<{hDo7NDoQ=zVt}~?m8uO=ҽYz~mՇ?zz!::OmNuOF{c[|Q[o.?Fo=}o_~ζ:}թǭWFooo}Nu={Mz_޸u[׺u{{^׺uת>yu\R?_ۃU~!]S˩k/w_NԸ{ptR}ǻ=kQl:n/ZZM=KyoΪz=z u-OTJOnּR)M=JAEΪ޽JSn}3ORs7Ԥ<}u:n R؏n{kpsR#o~GRCx}ql3ԅ>:cv>]W`~TOGRGnGYzYn uR:έ>u^{>kau[ֺ¤uWjy=xg%޳׿.}Ԟ6n?s߽yS2>OVGJuu'Տ}O[afz=zt'GsQlNFlcl`:>]Gs8Gs=OW^7պmFz{㨎~zQ\lקQXmߗQd}?ׇQ_"[g=CoϺ7WD ~I}:WzlQd6/mSӝCϺQ\9V^ xu~V$ëFoEooϨlq3}l9W}Á3#gX_#Dou>[ˬ[^:;Շuo>Ot??}=ӭb68+}^׺u{{^׺uЩqۏRۃRwgMu1~8^G}}:l)?>kHTn{qzR7R>zԤ>:ԥ>H}1׼MJEpT-OۃFx}{}Jmԕ?➴k)ǻWuOV 5wSkո1S׸m=ׇ[};=]GQ}>zW7>zXmǫ)>ǫug}GᄒQk}ѺpuC]\u϶[kF?=xt3ϟ[#}}zpp,[Dcm:DoΡz|Qd}F=xq#}1ޢ9}o.9lu3V}O6W;}zQ$?ſ}Ӈg}S>իy?_zw{lQOW;l ?:OV=Go`oCՏQ^l_tG>zX{G[b>X#޺\}]{{^׺u{{^ѩ<:zŲ:í)u=qꇏRPtuCԵ<_nWJy꧇R=WSqSI'׷AgǧUcH}:l)ZOYCR߻#Ϭ矯ZԄOԀx}GYTϪ(>:y`g7-~~{oZ&8apz:%_)`zE׏{&>S֩(&?_g^w}z%~z:qOk_o}jcE~OZc#_ S/#_Nf{=[-#_mz\~OǺ:/}ޭOufu:>ϧQ=:ΣH>:ǃ>cӝDsEՇ}omEcg϶. ~ެ0:%#4z lzQz=[aW<{l˫]Dm^9{le3}zXO˨rt=_}=z[#Dot'X<:g?OzaǨ?W^7Ag;u8>uq}?u={ggϭ|}~,u^׺u{{^׺uҩ#'{:pR}ǷURndD~o:qZI?7S }cu)?^zR?xowM.0?muîAy{pM23`_Uí=KŹ}} z~=xۏ}߭!M:RZ@{[GaH됐 =S*U5gV'ۜzS/u |}t$k}8^9Lq8u^=ze TNa=sOI?-ן{>jesW=r WuW^_~^\/~/?ׯuUpgOZu:u?Ƚɫ?np'߉|:[ߺ׸ŏ7a`:a}Gf}>`)>OOQپOt=8: 'yk]`caՇQo>zXNmn?'#7o5+on}C7_^ǨF=_:t93mPd[95=[-?~cӃQX???m[DsϺ]z!WGQ}OV+m:!}ot=lp+}=l?l>?Y'>[guOk[޺QFQ?1Qlcպ:3{OQNX=z+g}#V`s_'[}"zˮgo=ϫ:^׽u~{ߺ^׽uөdΑì]b}OTK?{plu.5onuO?oCéHָ}HEoGIGqqsoMu=?~U?No}>]PlGңooǻCsOZ<:n.kuAX_ۃgm^DnoǥqzX_wǮA}n}:YĄkuB)dzOn#zΎO?׺ΧOM=HG|OR+ԅoZ[R:~{uZk=SR{_(}ڽx\{U:v ֺ$'_>^K!a^Eϻ׫Ns $?V,G{ [zZ&}׭Svyo=tqeoͽn:e }?{z pG#Rzzk_5SJA&?_t'S2"^P7foRz`f{: gwoՆ:]'~ОXSWW>"}ǭxOu:sb2q}ޝ8:qouר~}^3f7}m)g?_[<)6c{7y# y}m>ޱ t'9N!y[#R/ǃǭ<>׺>^q}cƽBqb۟=:ÏxBz3ǶW^?lԯ<{# A{6:puu?"8:~՗QV}=[qՇQ ~}GcN aomׇ޺XՇ=[Gpom:﾿>u׺{{^׺u uߺ_ԩ%pu.1%9Ti?ӏ:-Emu1j.>\Hx{M=IIOֺ5? އGϪɷsTq?_1é}On `losǻb>#)Ը~}:m<=iG]+z[y+^rޯonR:z?v^=W /U#\~?uCԟoqOZ:8vDWn-op>uf'v~ާRw{U냹IfqpsXu1^~G2>Χ\ׯR˥I'zd/{N׃p,x^#}7Dg$Sn\u'^RCG"6BzQgW1'sNS `u_?Riu}p?@CǫQ#BzXٮ~ou+|zpxs?>xn=Gviu:_7cǶN1ȿ? |Qm6zNQg?_mgag=CK{Cg?okq~W?p}u~9:4">}__ɯ[75?ot>]8:MÅzgϫ(+߶W 6t'[/ч}zPd6[:OmzU _<{N|!yuǨt31=z{[=cǏպ`cQNuחcc?~-Հ+o.~{=ҽo%bO^-7CÇVc&q\:?u>}oׯu~z^~u[^~{_ƽuը7{a.Ը~sqz׷u)Upt\a}ݺB8_<~}:mۋz?{qz:Ň׏?>0zl盞o{rjgYt/r?_ngjz?Q{#=ZaӂOǏ\%o`@x_. o=L{} }{DW=LY?:_SUOAxM=ZNPH<=TH[ȵ T^:݇U#Ӭ۟wO7?=ۭ =R`O{˪NQ?n72:S7/{{={Q|vTN8Ϳ{qSϩ|Cj[R>=xu~ooz74~ߟzYzNj{hEc~z;mv7ou=[˨nyGot=8~ޣğuc@??u=o狀:W'O[Gup6$܁OϺ<$s`?[%sߟt={׺Q޽u?u׿u{{^׺֨و$syx2"^W8w?pW>GNߓ؏w6pz*~۟Ϸ:HX59=JZSo~7! (}{9c f/>6HWPG#?^8u~~?NGY{s{[۵=9i&N=z=?SԆ}?Gۃ(zuqa?hG}i\tGTeGs>^ERQ5zR{|NS_r=\WS5[O׷ízLjzY?>?u+fܛ??テuRq2pA?׻U\uzA ~O{T]8S+?ou+ǿuZ9wͬ?S{RJ~qztq?"c[]^A"ǏsǿzuuT6n?#ޫ=sԏ=Ł?~zFl"_}8={'Q>^>Kr>~_>8X7c:GQNm ezI%[I}0Hɷ?x'VϨ#^ob=ֽ_-)6~}{F_a<+&G^),,O?>'7=]G]y::oozNBi^D[ߟ=_?^О2zlEs"ֿ>~ެ3V>=X$I'}O}О1W~_k :!ScKӜ:$[q׽:nOF=88uE&b?ֿ}XtQ -ոM^?{=9uB?N8u~ՇQ?ߏm:&[>zQe?=ǫ?ֿ"_OW^R=??V#SՏ u}z-AP?Oឬ:ss+[=Eby{oӧuОxW޺:}Ѹ[`c?[::}?:XϺ2:Ǥ~~m#kO^돽u{{^׺u{qR?ǿuרI+'?qˮP{pp:%\_w=P絁a?>oUakQ! $}t\} m^Ǒ~}_q)2 xӎ}OUoc"ӭI#S~9_=7ԥkwG=HSkg*\wT: O+=on>];:oWǗZZ}?EΫ|qO^doOwԀs{[{=WIk }[_b<},E 8B)B91 n>wu'?{]yj~?lGPj0q?{>S8_M:/<Ԗk}?xi!$mGZzӓoǻWW^f[~ZC{zzS?;M>^ˮ0-;o=WϯWgi m_}ӯW Ǻ:g^>uy{zS[ROUzzs.Rz^4/vz<+#OOu$ӯ^=HYE >ZuX?؏Riպ&RR,AՀ!Ô$o?n~O={)Kǭ}ǓXگoM7{ZzqBGH5u乽??cLB-qOt'^1 4ìd?l ma=YGQ>MqHykOY}>cӫ96OmB?}Ѽa y~?>)SN[0q[ot':OO:;ak|ϟVQMD?R?ޭOWXI~ga7?゚/Cy tcOKtQdqzPmr>oQ"ǟ?Նzn=kʞ_{|OXW=_JV\~x^:g:v?^GaGP~a_^Wycu~{}c+s垽׊'>=luOϽuqgoN>{}aC׺u{{^׺uS}|0WsϷGT9Y==:nEMsonNEzv\ȹGNԩ8Sd<<^C }u^%TNAq&y-fS[݁`h)d}e\./[ wZ"J UBQtM`xuB:.A6K>:M6Mc}ڵǯu _SsݾδF:|2(_v/{kֺ GGZ|5}jl ~IzONH~?uB?O{Zf֬%Y X=qTX(^ִ߫u?r~'iy}.m>Q o4@V1Gќ[>^өeމt5[1!x}螽Nfz;#JS?}ԟ.w};:?'[t'?P^:Ԃ>>Nzu]<Aot=8zslm>$}К}I ŏ~on}~3#?{Gc{lzW. #}ҴEp?ןtWnHo6ǫO5}Gkщ@qbn~|&*tx{W=gY:u̵GǺӣXl:g?F=XqXH'|q?o{WV^0{u~$N/#{l~H>'ոA8cS)nG}qì28_xu W[Ot=;w>ׇQ[Gu~S#u~?{lt/S~={}z]Fb[~zX׺ߺ^׽u8FO{"IEO^%i8׺nx~ѧ+@ ~o8XGm6{puԹ2v['?܏UZs&ߑO\sy ԸRiЌo?_֎GNO,BȽ):eWwTnm-wg/{puDWHo}IGǻ1Ԑu=SRQ{T>IVZ9fmowu\{ۀ**^nykDW(yXT^!PaWz95?['݁=n }~T#i Q@z׭t b G?`iֺʲvZ\K[OG\Aھ}Tu2G^NQ?m=ϯu%}hȤ]@'WPW:G[Tڃ~רzIEO+/Q?Lq/:;J꿳9!L*@'_6󛟨S{OOJK[u+ t&9޿])% 6'x{zJ@ƚ*9$ޭUt,a"ia{_ՂǐJTk7XKztu Eo}O?I}J{׺̯W*YxA"ߑ,?7_˫R"fVBC]JOu~o?ӭzBzQݯaV>B>&OVu{ ')ot'z?>ot'럐~]o l>~s`yCӀPuŘ/?{l p#oN:X[am32y{9}Dvo~oWGbIc~=OV$ yXo}}=lqG67?{lt1AZP>aגr##k{lNKI}!`IìN=]GLՔRg^2 ?(}ՏVXٍ-}^8uqEnI'OVEa~?޿W1XП^zo#goVEsNd4gȟGNV=Af CՔu}:ߟluOԟ_G]zcs}=fP[_}=]qozI~PM'GtKIW1[c[~?}U%q9-?=6$Dy6 ݫ֩׌D~}W^1,1Oukg~uҙ$!EG^MM?j>zr?OkՇQlx#'ʈ4ʛ|uzMW덀y'^ack^<Nu=oL9^q}ֿ|/?խ`?zXm~~q:%{}{޺YbI$jXnu)`^-)o~ewҿ@{>׺Gu8ߞ=uߺ_Ӥ6#oryf ? :8}±d}G(hOL9 'ۃת\-{w|l$O x{=QMRTI+ݝ\۟wuȷ?_>1zKY&=GNңQ"ꧥ 6^Uɷq)(mqo;Źpqֺf v_`z4 ??ê\u%ߏǻ֩ԔxzFJV_Gqpqԅo>1Yz9r3>ޤ(?K֏˩ kz?#`Ը MKk}=TN.ok}MbO}W2M?W>6E@@TSqgQ_}?:bcbmuu3v$}9~oUV3#ZJ8u:+bnV^-UשNƻ8/}?=t4uj.i3OK>}ҽl TޫՀ!&~xWfÑǺO[2GOUz_^5lߑ竁W%F}9o.p`rBz6X޿OVQZ<{=X!#}ǺucS?B/{=mGX \>OWun'c?/`mX uѐpx#_ §:`x<}^qc@>?{=oϬ,m{?=яNNjy??=ОYYE_}X.XQE[oK:'-aqb?7$}О=Xp ܱM7t'I,5Z߁}\u|{ơXu:ǃ<[t=X#s>9c'i ~OX"[:am?ӃaoU珧?>qՀ}"߂yW;7 }mϫpGeOg^ߞ>M}Ԟ'R _n}V{ژ HՇ\ySu:Wgcya]_,G$C>{ 8P߃o}}:?zX}ӯu>I~_?(^^8E2g2I7&}u׿u{{^׺TA${Ԥa'$XUusr!=+jG!(wPt6FI'cz}x}SEo?TXKX}xpzoB)>_~]T.([/>U9<5fkqϻO^ӑ ?u]b- >S׺ Q qo=u*t?ZO~3Qě kO~b򴒥,_VB_y}HV,4` 7卽꾽{ԎE-~MI{˯?$X_މz:P{֣[NبY#3,د/ q~ Tt~5}`z>שIS>OWNIEXtzRKG }?ua겥={s@5 Iu鐞8ׁBzu?C~A6< |>ޣo8C'[=cRon[ONp_t?o[XnGo6/>O`u?>ӀS:ʋ}O {5=o!^IOt'ӫ@6\t'V`6<_~zQ[p~pURzpuB?}_}XfBmD_i\uΟ7QӃ} _k>ǫ{o[<_# p24Xqg}TXzǶϟV:Gӟ-Ǻya.`:oz\PֿQ]{ u0?z'I(yazā{>z]x]~?}XpÏ}\=XuY[_~]qZyfaIɷ$u={qpҁ%QKT\z^FvE/o~`[o׿uouߺ^׽u~{գo\}XLt%vT ǻGY'G=f}s"-kz̴&o}9`|(z(H_Xu"<0i*~$NG>]*i1hTӓ'{q8TݝGz;GFkQ}}S cz8~>q׺SYǞЏT#ϤvJƫ~mu^)y;j>G?oNm9ki?H,m~hԅ>ZǬb:?dYͅ߻־]s7`. UI_su^jX(~=JӍ.zK)'ߝ*~׭qIvaQ Z Q[ܗT6(p>=hGPV1`6؁^O>Tj1@EqitqKjѕ{nO~`uq!rx>nd&8SH?Q`׷ש/ܪ:R Qk?%S>z+T? G&sOVzJUHltܙ?"[n}UxunSBf*Hڼ}BU{[܉M2zeKŘxz?oO/}o#=ҽoR~qzBH[7c<:FD"vN#Ԑ6? [ ^X~n~GԂ:$\o 'P`G G!>Xzu byklup?oQ(?}y?>][ڹSt[^>}Оon?=O^>Nm{u80?o{n[OV:ͫnyBz_ A~O$}n=C,y$[n6[q~}\Hn@"ËҧΠ- QBqf "_m׫ӌ3 GϿu=Goì̡&^'@xm޼V:J̶?]lg=asom=[%?Rx`)<zߟP@y[}}o:ez{}7)-s_c :if,<}K?׫t-IP~~}vK׾}b2n'?[`ru_Op:sSgeSLp?ߺIO*V9>zMҒSIOza"{^ߺ^׽u~{ߺ^r}tQQc`ߺ_֣ep kͮH7ϽGRDΩO.G_:oZ'B}OT#Ip=_HS3/}'݇ZǛ~O߁FMLGq=T+?{9d36?ۏn^E zs=W${Z{pcu MsΨFz}>zkϻGYŽ=fG|{zHV}8=fW}}UC~k{=Wֽe%Q;P.O^pz=:QRAE!a ~F'X^EzAl 5+&?QŅkS$-O{^鎣'#(t^ͮ.=W^鵚G X7l^^]J=B[{zJlhӄZ_4WuՔ+($V,̷C{4 3-~=pK\ނ4-8bzLAd,8މ={FbI \?z?ϯ~]gӵ\z̊n{s͇WP[pxx~}Z5TXfn5NOq׭ӧ&0\T>Pm6׀JǞ8\ͽՀ9SGiOϭӤǑ2tA,Q{=l/9MzS?: S?RzFP|? azK2c$I#K=_QK HfQszb,rc#~~M1XUd[S?=ҽ]GXϭ%oh!=КV[r$m~`tmmGXd-??}:7:$z[OoNB?G!ߺP+h=?{`buk*ăgK~޽Ը85HE?n=ϫzz$,\Dn/f޺Y'2zTi@=*,8B7[7>׺{,<}uߺ]ێOu{퇿u~aK)?O>׺ai \a{ע@>kf27Ԁ>Uu%\QP:zUp-`?װ@pSu29m=ߪG"_zaNJ,>< =4qԔA.?zHZRowTDOoMԤ|nGȽłv?ow˭}1TN_|HHF폻zmk}EzC=P?=ߪҽd {ozq#Qvrt67U+jZKW<ӡ3DL1-[g]M}OG:+ç\[68_oJ,;ìjDtƢo=OA h Ҡz>^Q9:ɌR_q{,w ,E_{Z#tz̳Ri͸C 7G=5zV?R?Sfc׺6RG:T^ܟ]hoDR=Ԃ׏{ZҞkT5B5>kOJH6@jUozz.&)1Flmoƒz#B!G:A2uu KQL]R) ?{^遤^=l uM^E}԰tH8_Wc8Hb,-a_I}aXUnZWl ~}lԬm??yVo#X5<.Ǟ=nz{pPFvzC=u1iREכ_U\(@W꽸qutGŅ׭Rc$O^:G[z=\/\k@=@Gl}Оt_YȖ ,n9<m'l^RUOsI|z`[(뽇t}l R.\܁UEzEȸ=П.P&]D^IUr.mazMrr/r-Su:IcucӃ:Qt=X2#p?uĶK_>A{X^Òy`9zvp ?KǺv${y7϶ɧV:X'Q?o{>z}CdInWnGq.Q${.m=ӏN|O%G#={mcAap-O#ӫ *EӄU^y}׫ X )?ӟMCǯS\{l{!BY~}aBo[-Ϻd?8}::đrx8C׺Gy[zXqu'uQ?Ԏ9[:ptn?_G{&]-nt8 Ngeb%q~l/ou'tX:$qn/=׫yIgf'OAϨS^?z^,-Ou_~c{?~{ߺXWߺXusu O4Q1kZ?Т ~n9g>c9us׺̲k}uR:?>ꇇRR܀nGSmG cOabe^,O{=P"g ;E_w0OZg\-#ۀNf|E%?#R~8nzp//~sZW$Y,XS͇{z)ӓj/}wG%؅ސO7#]O02FH-Ņ>^E:xu p?pGZ=F7* _ֈ=ɬpNc#E>T"Ayz^)$"2-~}ڽhuZ#U@\Z=M=3auiҢv†V"`s־ޞ}TǑuɷ&:X3l>J+$*oi,Gđ`z+æ}Yډ;il,{z_>j%jGRj:n{z+S3f1[*0Fa{^@|_&' J&9[[o$}?שԩfddj_T=rܴq4@PHשtܱSS&'Fqr>Sֺw o~[H}W U]mR#s,A~]#^m7~^9~-UF2yz:o;9mzU,Qr@ہϽWS]To^I}/~lHO^LbƩ6EzޚQ=LuÍ@{4z^-CڔqȿG׭ǧ(+ؑrI㛛oމt^NZ2O#Sۛy/p:mX,?F7$^Qdk}Oͽ?Oqt=\ W_sո \'{VX?_t&XfT$k܎-ZոHXiMV?B[kAՇٞ0p}Om6Ň>պKrG<~9_~[ $/c BdAȟM_Հ?^\uaX~[u5gӧ8N,?~s[zrʣ?}XVs5׭U H[n/c7Ou'<@MAϺ^u"պ GVO?pzBk׺m#oq׺c#~_t'} ž~~}coսz\կ}?￯uZ؟~p*?^׺o~p7~Gr~/=utE0[K-׺ѣy?6XEz@[ݺo(߾}o"_v=6GY۝V$8i}qkPD%ʃ4Du+,ʡ}@NzS1_wBY,IUe)\_>G(E{q:=E CFroͽuR+pQK#kbJ}~1B(z2i4$䆸OֺqV7=Op?mǻ׺VScPoMqdUww C>U+PIQSwoowSQ[IXG{GYc~?W%?onON={dYy[”FzsuzuI}~ᄋ}_>;~?zTX6y`zN4tu54Cf:A >^ڊL1OW'4ִ&Zze P-ǽ]=?&qBmkO~}jӐI[Yk^\ŠS4LC%d[OoϿWTu=ȓ3N_^iX_?m]&**޶ _{zXyz}qX]?O?ޫ׿. J>^t+m׷WP[W.Lң~'T&~V :~ƔCITA#ȿ޵uﳨ'?AÀ}nq19>qOt-ױj#7$eS7'މJX8W(9 ?O:^ <ێH<z'p:\-?[Oשaȱӫu:}9{oh+Ë ^f oacRzQ,W?Ǜ=lՔ6!%~n}~r/~ZҽofIP)nC?-csS\u*}Aaz?׏zcQY},=#>=XgsҞtvb} {z$?K'Q~Ou]GOBzGi9O7>Շ\n,z\ Z_o=o'mqo~GJ}ӫ u~ N7>Ӄu[x{[# ?Aby #ǺՀu=o2ʷ_':O(z*v# u'7?_>Ӏcrx?_'ޏ[kN={˨n,~މ6J>oߟVG,n9x{VuI?׏?z^@;?^~uo{A?>׺ߺX%k_<_u1o7SS^ҥ 6m#}'%aܭzAd#[]~}y: 7:S`-CYc#_N}zZQ5no  pzϻu+ԘOVc~9{Te0K5t ݁Ncկ^Ky?C?U#}F`}zGYU7?OݸgPH/c{Ez` 7[ǻ־޳(}ϽEzĩyDIqȷTNQEY't KUYxĂm#utUUhSv.u{>}ra3;j(կQjk>}-Y%YW WԌ,S5$lDCqo}{8P_?=zDr>Q^A7-ZO7Ζ:o`oZ@AЈyf~uEk_^=lqyqs=zZbQꎽNÂۓ}:cl~^I0oǽWug=;`WFm_u!pk>Zk׾VTU1G~<{zRx" zDL9U'{z^ǺX EQsot&[ 7HԞ%e*-Hf6^M6O[#,t)Q͎3U{Wb+v5:llxzqHVlnMqǽվ]6J"bٿJ_Y_VUtH ŸZJdC`t2o^J?{cPWOB_z]WODmǫp/z,9N=x=W.oasBiպXV66`>rxp:XEb?ࣂOu1oɿ\N=mGQ,?BjGN;j7_.u#,Dr??Ϻ*g@T?C6=\u$^\zg:CիӤS-tK[O}4H%9X{^1ĀOպERR5$?u X?ǽunO=ӁozzQ?Q:2)O^æSA_9y˭Oss?_ur>^*M޽uޠ??^׺Pn/o׺P#ߺX~bֽu6>o_~׺aoӨXێm5<)6>?؟׋qQmG>ֺ9~ֈ:HAV"NxmzNg[z }ڽT>!~}l__U &ڈ_6Z/[Cӥ.VoatPcU8HEڮ>O߀O?^E:mAdzMozHd{#T phCN?qS[!i$v$'y?Cu^#ҖHZ_#>U#F=ڽT.Zcx^s݆zԡu=Vsx6oắ]^J?δr|q-ߓ֩ӴoO=tnϏU>o{Wצ5K6?wZ==a%e"뎴W #?[߫tϨC~zP+#.ِH ޡz:mEP~'Qױ7(zK!o7\zMtBLӟ-L7E?{z^z{qǭu.%~}z">ֿכWcAbAoŽ{sE=K4-B )=jJI(e,VUE[kzޓҍ{hѕp{?.vOvV;0IIމrgf`>몽n/6+d<]C?b=ԞBxČO^Lو~o{R=־]o:ЄOEZ}ԞNJ 8׽tJ Eu<@~VR߳#ooпָ[N3_>M::$㽽H^O[:oڸy:VO:~#e>YdcpZk[ot'TC"!Cr$6,m{`+.5;Tқzmsp/x^0,9_T}O?>\]E"Z #M{>_&Ŝ [k h`zf<{<(zij$?7U}>Ϫ䏡}l?ԏz[qmup:Հ1y?^#Ig: mcm`>OVRj?Ao-@ROWϫt2z].u^ëpEFqn]Z Cp\pnOSӣzr~^{֞Ik]_%Iot'24J7n~}ҽ8:DTӴNEOo|MVF=׭!YJK#^,\ }ӫyTXٹao~=[ ߏu=oRFPׯu>׺$?Qy/~{ߺ\?K{*x+ߺ^n~XߺYB,/t' m{7JԪ)4Xtnma͏}ߪӠ"??pq֏Mַ|QXl9kZ뒷?~U#aΠ?k=޸fYЛ`z:poZ92pHbWq'Y?ݫHlu?=kzOz̲HV$$pzǻ׺%A} 6NOOuz5[dd.5﫭S-uR&]$-Ͻ>6]\]Is=j,i"@,>^CӀֽ}{^ֱ RʈǽׯP]鋣 Nlb5aEz $̑7@1}{OQh pfb^V<?z֞Mң~=>8UĄߛx֎|lc>'~[Үm@J%"ީO{zގa$%XxՀzޞLIKDIxKk,WPq'2:ڿo^*d'g/.ɷO^31oHt؏<<T}x܋SGNu56qscދuWfQp_Ԟ56MUsi hؼJN:6ù)] -.:JO+#Ե}>_ۑ[/Z\ybRGs1us>[ӏV:s ~?K}П3MP7RzXQGcV R}:룗#MkH}OX_Xz.A?>O[Y-[3^$9][s:`P^j_t'N*#cSE^I<}zjM X~?k==Z:K_qƣ u\~ R?Wsen uk6O͇:GnB?AO:9-}\Qp?պhP n>q􉬦xZ ai׺ee?t=luH,m?"W7!}?]or[oz'uܓ? ׺"AKNy{D?{[k_۟~x9$K^'u/ߺ\{C?IjzEJ),ass׺n=ٮO~ժJgGԛ=cqNSwZ>yK]/^?_vUHYսs23gۀ׏U#Q>HRۖk5_#Ӎ&:IqX>WtBWWn."݁1JƍOH2ϨXvǯSӥ Ljঊ; Qc-4꿗]oڕ()H]IgeR:꟰k"IrM tP5O l}ᙑU\k ßSֺa,pu!/w^-] > nNn#zB:)KO_y_ӭtX$nMa#E)n4BnE޽j2O=+A}j)nJTC_=V}ޫ^Fi^O^r̪lO\`^RVI[~yދSNj)i2LGׅ <u&[ˡb3H#l?{>{d.O_k] }mA{=X^JK~I=[6MPF<~I'~ΓƼ̓O܏ׯqԘ2W syW ҫkf`Ar}vEG>OVçA-=Rܷ직=ac_u'u9BhÖ?=N:Q]C_iV*,-Lש^2~F39#.>}[yʣX?Кp+TE_zT+5#?AϿu~>,>Hu޿6OdnO[- aN=`5xCLJ^=xPMaf{ ~zQˋBsczbI><[ =ԞzEl8Cas^ .fKzݮM-:Qw,{[\5 V(m+Uӥ-iR,{?:(scN}ӭhK??aԞ:>l>tz5?K{E,o}p6#-{,-{{^<6X{^ߺ\ƿm_~d^@|=u_[{^P.6uZƞS~kǿu f]GTY$S1H?u֪?_yp#i8/s8u}DH$/M2~ԣi?~oGX[5}W}Z"a`窑Uk{ppIVy^6N,pxǥ5 $ m䝿Kk{pGS^le飦Wq΢H^ֈLTsŽÏQ M?7DWJI7$ BN $@,Omr Wqg6cOwh=A#6uS[UAIDf򺭕Yҿ_oz:PRmIZ FME-iB czR:wZ* Bͮ5 `\㩵8C[Z#ϯt Q$41S{/ӯu0oW݁!+OZ{z֞[f+ս=jks{z\vs}?|{z8JM_a-gZ u b~ϯiZq)7 \s֨zRb6fc _?>PQh"=@ 3?=W^IȱIO]7XRO*WNӹd *I>THcuQѼ^ڹX]{Q%j咞IYⓇ-oǽj{A Zb5)@M?^j-MQ䑾m߳D[rX+qz:gkHIzޞxO[}JxtFqӭ'R)c'U?__ޫֺELg+B RzH|䵉6ۓzk0Hy:u'SG`Cso<_tt( [q޽ЁKPr!EGP׺O>u`+ǯ^mnIOn\|XJW1H k_z KN֜W&_PUɲ䟥[މ\/\f?`9t=o򂱑ƗӤruO.ddjZNGǺNO(s)4Pt<^I`) _}58ۯ[MNAV~EDz8c0fV_t'|$#L(^\5*ܷUMc>?=괯VL?{=^]ag 9$_zp/Ф[H} 8gO_9OV`/=z!}LmcơXsoWcf&}gc=DpZ>螜G}*msMq:(!#iÇHbO_~}7j*Ãӏu7o}n=uˏm?_~au#}u8OoA~}ulW}ur>@u$-ۓ{K͵q{U͢yu"'7cO~) Wx׺Z&FU D .R ~ש,mRD{s>3pEAvhakqk=Wze77yֈbfOk r/[P\Hmֺά8OU#[ۀת-S*FZPE#H^ki6?}Ǧ1޲, 'ֈc=۪tKF`8`z@%A݁zQ00RTT,}uB,MjTAP~Q?_w ~ޥRRR4E&"*Ou~Υ"\n@?OP)ic2T:(AET#{KS zLE9{=T6Gb117'c`zRC<{^ ~?_=zjNbM{i{\_{/{^\_Tq8.KZ'K^"$nۀmꎽ׫sOximOOu{w`n5^?u]`㡐ԥHU kG/^JSSe&e">>}m^NdYKj?-=PuJEM'H:@{AϿT~}zLGisU#DrIiakoz->lf%.mq>O[ \DQ_\,/o޵yuu={46ka?Oz:H5{??[u'6KL o%lAO[Qi8(qso[&ascVoF,xԟ^ yO=oX^Ou+]TP?OӔ@fʿ?Ot'dВ@SkrO#z֞@X\[zot-S-ܛF "Wptj>W@ `´ |( ?4tt]!\S:$nDBPn{W;hE J !)" )z=X' O1 m 5\[xu:[ }M~[WY MО\un?=MW?}uu^Gј5ϽՆ3".KN[rjcI8n [TWj\ܐϿuS,X}M~RzIJqG{-׺Ȓ~?ֿu^mnm=u???oo׺.PIu~sğ~G~>p}tܟoߺNGߺOL?kJA^׺Ф`ȬeWr#ϡf&YH7Ry/WMZ#DC~G=WzG1Tv o?{z^G6coҝWq[ gT#-s][}u 7H]=LowܕhF i$?>֩>y'{`z`@/{ֈ-Lm{5k&N$_bߛOUJj<=־ޞ'F<|}}kjRzQL7?B:H x苟B\uo!vjn|JX("_z\<_I a`q}ڽT\?ߺ\+ׯ?|^$9<q׺-?q36ǿj)L'"Ixm]{Hy)٘A=}kOOxX[~ߞ:֞4;y.@'~=I٘D),5[F,Ez((,EtOר}:B }^'1V/"O'ׯSzWuEJ\)?MzXX`QG갻_ޫ Fn>E{OQF {֯[ԈC#ƽl#}T~-ש&_.GtVAF%,u-؋d#{>}{!b Ф"}ӫmN'~l 8n8ZujS}ac~ӏzzX̌O+E=[mpH,}Оu~~z'ҬQ©VäsԹYq'=ОI5[$qO[ _K~~}ОO5E_vsZ:bFqƺ=^WP#U}zqyqBUCn=ҝ[,G$sk-qrEuduۈGԑz1iHiaboߺX\g!2oV>^Q 7G?'rU}Ԟ=_0({>Նzj5?4$R|%R r? O^aAQ6t&[o[_OV+_nGN<u=xG~ou$W?F~(?{#u~\G:j6xRz QlX{V@O[OՀo6u:[sa~~zPgBH\OWȿ ?}zYNB=`:&['n?Vž?~?CT$P?{}bHm]{U1Rs{셾\}8$Ou($s_~J㟯uʦߋ?{?C[_~a[#B@[~c ?B??){40!F9<~/_~=&(  o~'7$#X+؝W[ߺ_ѡ8&zy~o?S2ǧBLlMoh*yRT s}=ӜMaoſH,ѬAQ6Z:+y Z}:G]}XےOO"_.FX `{uRc5KsK=S#ITl,.pzN0-L?iaae8ګ["=jJ;0Q?j-K맕,oꍖ_[ =hP)b><?A}S '?C?Ͻ׭SZ͇<}Go{]b:d=z:R P$ck_Ͻut%AU.F_CJQl_^)4CBO^4ټdU?u?~^r|⩮!-;Ŭؑ~=W[H,Դјc$GbϿWtW]Di $oG+PH7S׺鲅D(/qWtE!D̚b?Wtش+O,ҭn9ZzJ\&* @orXZ:Ql~BE13*G4AUCO^ybzyM6#Xoͽ]{.fvz76޺3=bn~ދutĽl9"ֽoV:HV4aǽuu>skn?7O[ߏy=tI_o׮znn}ִWq2x-#I=!7.~8Dv_c7H/ܷHoԊn nBz$ osO'k׾Cդye5G茛z݇+[סf$^תpb5ž{^ԕk4:Η#݁X=g TzG{>:`j}1coEVkhѡ?ǟu4iUbb*oZå|XIL'a/Vosk{> -Y5O>^W$n  Pݫxu>>O3kʆ_vqH(֥< `z:wBC*Ñ"߁DtHO<\G}z;:lEŭubկZ#jIߏWD ?{ߗC觝,vOWP9+E0c>Zu:0^#Ꞵ:0, Aꞽ~}o3z`,,yoo~]cjݍǿTb!!EkZ>C_Ͽux04EPI?Pokr^Jis𱉕l@]z.b>Vȭg7 }oIMȥAbG8ꎽ/lO}=P#Q㮨[s#An/RzNZgk 7#}~=ꧭ#6=zӧvIZ@m?^O^ZHb(?޿{~>MׯP O^Aɸn-}ϽWP.-pGRzc^M_u:B P[Zunz骦y$v,ēӟItQmo x?NZXy<Rz+|@"F-ɰOV:1tCA X됋:M͸O^Sxjz+jrןtx`:-,f649v#I{VLur4r ۝O{[#yq߅-ӁOWc0{GOW9,N$L(yf=׭)-m%IuR=Ԟè!vػ`/UG a>+ǯi@"(hcG>MzM3jgiƳ?G]l 28 #>OV>77'Q`M?~zp.Đ<ozWb~=ӯq={^y0,{VeA?O[ K_m{=Xu{պajq>O[j^}GׯypP.O'RzQQ}׺0 >׺ kOZmiW6?_~(* ج+ pA^Ӡ(?>wGQ_Pi)P^"{_oC%Lh~sh5_?ӪU#e2-{owz=RW߻fT>Fa{44F6'^G\?O۪u Z*NF??vB)ԸɿAa`k֩,sQj:qK_ gcQ_kL+GƨP,Y:WUCǤo?_wfUS8O}>4jk}Gq{~}T'XG<WpS̪G%9+*CG7zR6eU5KGP.mC0x<{z:{~}x0ߵ|^$-ou={ȱ6OǽX7{שy4?SL^",Xnz]z/2(OSߵu|IjY#!c~mZ:Y5P~TjFSb}>}Sj`PkzI?C%j>=42JFK/ȿu.Ǐ׺h%*oJno{zޟ>045?Un5TRn9<s}W[48{ ym}{=?ӬJaz e͸ӏzV wkIt[{{pG&ֽn7䡦RϿWg7cIF?]?_t-,LrO%kթNBz}:Z}|',?I?mDHK))^#?: 1 diAz\eӦK_էQw޽ojv֔yXcnJ BGߺ\@Xf$ Mơm{>}XK4.#@EQ_?D{5rD *eWSޏt\X&YE$OիzBrOz`,̶MJ#y8`=I=[aߞWʽl@HQ~OGn6'X[ğA6?Ou>xfZC!@ ~ONN?_ߺWq=Gf?R,?zռ],_?z'kp~O[on/ϣL~/?>O[zrI~\JqOO^3yߥ~la&_?'ߺ'es6o޾޷i# \_u'ux#'׺?^׺HX>׺IVߟ~=S'8~?}uDYxǿui@׺{^뚵?O~K >׺"s{S{׺ߺ\GG??^8~~9Rj㕅I#׺Zi;%-{|zڣ`q9zaDtU SҝzLuq*[݇Z#P{^9{zX=g^+Rekn ?Qgػ(͏jμ}GH sۏϻN?7Z"r.@\~?vP #߿u#s{=Ok~sc:Q!4ߏ=jvj޽_^ FP[WׁDztDl>~-ŀQMG򈐽r <kH0+om{UUDQFYP{z}z 駇\T^Bl4~sG^>]3W!I͇ӋOvizޫ^jEU=x8\_?{ޡZ]ku~?>^@s %ҧ~~OO\!iz5ϿW[XTѨ _߫שy,~?}Mޫ^TT .9{^\!CPR?>^A (U!GsWcTXX^{Qt4t@~=AF?u0+5u[=* hXq߫:^*_u'z^9>^A]v?{}::b? թi]ESVQoޓ8>5y>ϯz3ixPs<)7o@Q!{=qdH@XvnzWӀt:X-NH u=6K7Uol u#H#oGoȪm.A?ZOV狎@z1׋ _uqo<^@Is{]H䑥ߺYcr}{׺ʒ~;Rl_~KtYm{׺d*oa~-ǿus=u'o׺\}#^ -[~ae~t ׺a=u{^)>ocHE?݁ `z+p^-?:l hїכpz}z1YH_X-a?ݫֈ߇Uj5DtYT~:.̮0<`z􍨧(H#~uB:T-S7u}YP?knIXpWG>߸uo J1~`uRn(_ϻU$Z?$qTӊR z)Ҧ%OUԠZߨ}?Zz'N^Tߒ>w^=r9K^SS`8{ӏ^zeg~n_azFtoO~[ ;VS"38}{H ͖T^:`ڕԏ$oz[("H* -z:In/>^!!&,xz 랮>O{=ntI6}b@~_=[鎷7 Bl=n"5'6mIIz}MsCw`yoozS@onI./J[rVq=ul?OGߺBׅIBw@[?Ҍhۘ/P~MrE{zq2;I<I=[L[fO&˨/?z\KT?aFyooz)ǯ=DU  Ÿ8e-ڤroaou'`:\)S!slN+-4Ec_QR yoȉ?25ck}@>O^Xz"^I<b~x)w{[r ?㞣H'l q鮦% }9b.z=od1>o>OV2Sῠ^1 ֹc\~}cussao}}V1}Ko<}Iu/y'=:~}~Հ %I7Znq6o=ԞS^-o {-?Oq׺"m׺'xk׺6~I {?O~dI~~?{Zjƣ~L`9?O^閦A?Ǟy^9}~׺Ȓ9~~~KF~mǿu2q{s=g,/u,_o>׺ז &0+^ e?y`^vFo݁zM e6/힨GBfF6E2ǻípC*Ћ?Dzu%z>'sׯ~7?>Jt/_ޕ)QJP?ӟ׭P2~>ӭuqph+}P:Y׺Γ5o}xBT86Gn>U#Y:%<~/{ z:oǻ}KD'z:֒n{ZX޸S&ױݚ/ Tp/o{ z:MFC6#r,-ںO{w?Wn mj?soc$SH/5t,@/>ר::,B%,y^F̾QI ׯP#6 8?^??߫֨z޿N={=p5'9 Xӟ~[ϟNTyH@!ȽWӭ~YJ2(:#/ߵggӥ4[Dꁿ gX$~>.WgJqMӷ<"\uv$m~Ŭl8$Oz^9I yv /{rn}oOKH`8=6m𽹿WׯPuY~}oa`?OϽjqyQ٬>XtQōZtM(k?@>k&g67<ܓ֮[Du 6}ozt#7'cދu3C/?S`:U?kNIu/Nm?S>׺N$j&ߋ?_g[zq`SCUb.-Dp:'亅7*׭S(4?[Zl ٨ Bu1k]{=[U 3ۆ\eGߺzI <}Et-GIxb#?/?m׺+3(iӯxz>}X^Yłi觋IRz f1?z[ n58RQl%p[n}־\4*?/}Ԝu!wߏ#qQY7u76/,I|Q/B ՇG ? }6r~[܏^æu~=uunC׷Gob.I?؞->׺$ULxߛVO^ y ן]{D.?6oҷ7׺m $~5{ʚn {->׺j)gOo~d?u-.~vS{^7?PMu礨CI?Fee/u^FM?O?{#h }zd_xuj@U_n2ORN86u#$$}~゙HEf$>^pVS^NZכOq]X5ēߟ{Aoǿun?4z{SMLj8[ݘtMBoZ#HyI$Rz\5Жo8Qq=X݀7ىbĞ>}mu^:T(w?=l'.=О׉!\YR}=xu͘"#SqNH4gGQ}ԞZ@uJu PEL roza!U9(VܞE}x;)7 ZWϺW+' ƶ'/x@k+u@_<׺m=#^[Okmr}:<}kXϺ׺ ֱ}׫׋Snkےxq=u7?ۋRz\*nMG& ?OuZOso~x~?O~v4mן׺y>m~8S9?c5YM{FC6~fR8o׺{}uox^▣U{= _~>ߺPpO$ {piQ2GnEַ'ߺSdžR.t&{DxߺSxP1Vu4T7'ߺ_9as.t y?NGNI/ӟpzRQ`?h9S2(7`zIY%$nB?{z6R9?znKJǭHu\S=u^M7^`j:2USzSE){{tnGXqi}knciNu6"?ߺSi䭐 %P~o{-?-OOQ׺vZA#?z:i+Z]Pmv0?כ{zKfZHtVKG<{OZON.?_F*($-׽ׯgrdi& ߫_I`V1sv~O'߫Kߏ?c5^ aomq,o^I^ҼEǽWSV2 _z^MWWX]ZJu_I'ttjGB?_z`z(j<3 5\R|OXSv?Ou F{cԓ:F`y$ /9'I~?~}[ WL@"K}Y$cOpJ'դck?OzW t@.ăV@ONï6[s t?kָ JO7E<}# ~ÁGոyuŎ 蠎?z'= ?_t'"V.=ij_W =i=׭e '{Oԛsou&{O3r8_}{UE[q_I(ۃzA*nUX{Vb@%o ^B/<^׃߄G忴~׺럯u}ߺ]3*nOq^?uO=G{^?_{^[A~ou:<~?^*={ .?ßן~HV\^׺]Rdi1p *z??O~0Wni:i؛<׺i9jxS{}doH7@l-{mkRM0\O~CtfdȿA{Rfr2ru_k{^#TM!~7~LA~[}?֩ӜR^?qHj9`zY}ۯc\k_R)vo\諤z/vе*~GӋ}-6GKiU76#=z7Hy+Go7??݁7k>U"4KZZrpux=F vU#q]wr?W׭RvDWq?R?z:0IJ8 ~]>KKM!l~?6~}bLz>s>׺t,?Zj()"[G_߸A~-`=Nouuz:Y]M$*G \:@~rW:LqbWty$ܛo8{ߺ:<}9{^Qu~~8ykG޿״?xA׵=1שhfSTzNkq.ܟ{|Y _3rE7=jpzz{FxNj[ޫzM33<jtֿϽWӨ-u:$c~}By {^)o7}o~Fx'?~t'7kHGV׀&Z"<RMzY__z`u6f`\\?K=o;]+Fn8t'qI=l.'qzWI:HƔ[2 =׭<)!G?ޫt?V[l@UgϤ\`#NBMzG#Cوzaҿ~Nm#@q~# >k׫"BA r2us ppu~4HU#ӯuF|h'[jWNXz(3&G4ٽ1߄[͉M:\ *)[zQPUbčl-}+k鿽u2m_n>O^,&Ɣo#Ǻ˫,tF,oSz.:A67׺n N'}׺*J["B_[Mo{$1_l׺P~ޔu̵A~tnno^_>׺ȿЛO~p) #6<ߺP3{'{^"0'$N=u5kR R= o{Z2ژߺX[ߺX=u׿u<>׺'x?{uJ( / 0DTimes New Roman<4|d0|Wo 0DArialNew Roman<4|d0|Wo 0h" DWingdingsRoman<4|d0|Wo 0hc  .  @n?" dd@  @@`` 2*t\)  %     #& // !  "#%(    O$$$R$ٶa|ջ.$c $0e0e     A@ AԔ Ԕ8c9     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S"@f3@8  H ʚ;2Nʚ;g4QdQd0pTppp@ <4!d!d 0,(5<4dddd 0,(5<4BdBd 0,(5g4=d=d0p^p@ pp?* Summer 2003O*= Summer 2003 =Intel Integrated Performance Primitives vs. Speech Libraries & Toolkits Math Inside & Outside@P6 06cAgenda Comparison Intel IPP 3.0 and speech libraries & toolkits Overview mathematical methods for speech processing General assessment of Intel IPP 3.0 Summary6 x  Acronyms  ,LP Linear Prediction RELP Residual Linear Prediction PLP Perceptual Linear Prediction AR Area Ratios or Autoregressive LSP Line Spectrum Pairs LSF Line Spectral Frequencies MFCC Mel-Frequency Cepstrum Coefficients MLSA Mel Log Spectral Approximation DCT Discrete Cosine Transform DTW Dynamic Time Warping SVD Single Value Decomposition VQ Vector Quantization RFC Rise/Fall/Connections HMM Hidden Markov Model ANN Artificial Neural Network EM Expectation/Maximization >Acronyms (continue)&, $, CMS Cepstral Mean Subtraction MLP Multi Layer Perception LDA Linear Discriminant Analysis QDA Quadratic Discriminant Analysis NLDA Non-Linear Discriminant Analysis SVM Support Vector Machine DWT Discrete Wavelet Transformation LAR Log Area Ratio PLAR Pseudo Log Area Ratio GMM Gaussian Mixture Model WFST Weighted Finite State Transducer CART Classification and Regression Trees HNM Harmonic plus Noise Modeling MBR Minimum Bayes Risk SR Speech Recognition TTS Text-To-Speech synthesisVd$ !  5  IPP vs. CMU Sphinx,LFeature processing LP Spectrum Cepstrum MEL: filter, cepstrum, filter bank PLP: filter, cepstrum, filter bank Language model Context-free grammar N-gram model Acoustic model based on HMM Each HMM state  set of Gaussian mixture HMM order HMM position HMM transition matrix Baum-Welch training\"j9#")          !IPP vs. CSLU Toolkit,wFeature processing Power spectral analysis (FFT) Linear predictive analysis (LPC) PLP Mel-scale cepstral analysis (MEL) Relative spectra filtering of log domain coefficients (RASTA) First order derivative (DELTA) Energy normalization Language model Word pronunciation Lexical trees Grammars Acoustic model based on HMM/ANN VQ initialisation EM training Viterbi decoding* /A%47* /  ( Y/ # IPP vs. Festival,Feature processing Power spectrum Tilt to RFC, RFC to Tilt, RFC to F0 LPC MEL LSF LP Reflection coefficients Fundamental frequency (pitch) Root mean square energy Language model N-gram model Context-free grammar Regular expressions CART trees WFST Acoustic model Viterbi decodingFZ6E  ( d $  IPP vs. ISIP , |Feature processing Derivative functions Spectrum Cepstrum Cross correlation Covariance matrix Covariance (Cholesky) Energy (Log, dB, RMS, Power) Filter bank Log Area Ratio (Kelly-Lochbaum) Autocorrelation (Durbin recursion, Leroux-Guegen) Lattice (Burg) Reflection coefficients Gaussian probability Acoustic & Language model (HMM) N-gram model Viterbi decoding Baum-Welch trainingPP P2PK3  # -  x  %    > Q ' IPP vs. MATLAB,Frequency Scale Conversion Mel scale Equivalent rectangular Bandwidths (ERB) Transforms FFT (real data) DCT (real data) Hartley (real data) Diagonalisation of two Hermitian matrices (LDA, IMELDA) Vector distance Euclidean Squared Euclidean Mahalanobis Itakura (AR, Power spectra) Itakura-Saito (AR, Power spectra) COSH (AR, Power spectra) Speech enhancement Martin spectral subtraction algorithmP2P PlPPPP&P (  L  W  & W  ( IPP vs. MATLAB (continue), %LPC analysis and transforms Area ratios Autoregressive or AR Power spectrum Cepstrum DCT Impulse response (IR) LSP LSF Reflection coefficients Unit-triangular matrix containing the AR coefficients Autocorrelation coefficients Expand formant bandwidths of LPC filter Warp cepstral (Mel, Linear)f s6( & ) IPP vs. MATLAB (continue), $Speech synthesis Rosenberg glottal model Liljencrants-Fant glottal model Speech Recognition Mel-cepstrum Mel-filter bank Cepstral & variances to power domain Gaussian Mixture Speech coding (ITU G.711) Linear PCM A-law Mu-law VQ using K-means algorithm VQ using the Linde-Buzo-Gray algorithm9TZ7R  B  )       -        * IPP vs. HTK , Feature processing LPC Spectral coefficients Cepstral coefficients Reflection coefficients Gaussian distribution K-means procedure PLP Autocorrelation coefficients Covariance matrix Mel-scale filter bank MFCC Third differential Energy VQ codebookf_J /IPP vs. HTK (continue) ,  Model adaptation Maximum Likelihood Linear Regression (MLLR) EM technique Bayesian adaptation or Maximum Aposteriori Approach (MAP) Acoustic & Language model based on HMM Grammar N-gram model Viterbi training Baum-Welch training Speech coding Linear PCM A-law Mu-lawt':, ;'  B 3   5Possible extension IPP 3.0(Model adaptation Maximum Likelihood Linear Regression (MLLR) Bayesian adaptation or Maximum Aposteriori Approach (MAP) Model evaluation Itakura (AR, Power spectra) Itakura-Saito (AR, Power spectra) COSH (AR, Power spectra) Speech synthesis Rosenberg glottal model Liljencrants-Fant glottal model Speech enhancement Martin spectral subtraction Speech coding VQ using K-means algorithm VQ using the Linde-Buzo-Graym Acoustic model based on HMM Baum-Welch training4PgPPWPP8PPPP9PPPgW8    9f g )      / (        7Speaker Characteristics(JFeature processing Preemphasize Cepstral Energy Cepstral Mean Subtraction (CMS) MFCC, LPCC, LFCC LPC (to Cepstral, to LSF) Residual Prediction Mel-cepstral Fundamental frequency (F0) LSF (Bark scale) RMS energy Levinson-Durbin recursion Covariance (Cholesky) Delta cepstral (Milner, High order) Pseudo Log Area Ratio (PLAR) DWT VQ8     WJ Acoustic model Distance Bhattacharya Euclidean DTW Viterbi decoding EM (Lloyd) K-means (Lloyd) PLP MLP Twin-output MLP LDA NLDA Generative models GMM HMM (Baum-Welch) Q 1  @Speech Processing(Feature processing LPC LSP F0 Levinson-Durbin recursion Tilt Gaussian Acoustic & Language model Baum-Welch training Viterbi decoding CART Statistical language modeling3H(# |Speech enhancement Speech Analysis Discrete Wigner Distribution DWT Pitch Determination Code Excited Linear Predictor (CELP)N#Z#%|ASpeech Recognition(Feature processing DCT MFCC Mel-frequency log energy coefficients (MFLEC) Subband (SB-MFCC) CMS Within Vector Filtered (WVF-MFCC) Robust Formant (RF) algorithm Split Levinson Algorithm (SLA) Vector Quantization VQ correlation Single VQ Joint VQj"7",J% Acoustic & Language model Viterbi decoding LDA QDA MLP PLP EM re-estimation Minimum Bayes Risk (MBR) Maximum Likelihood Estimation (MLE) NN (Elman predictive) HMM (Baum-Welch) GMM Buried Markov Model Decision tree state clustering WFST Dynamic Bayesian NetworksdBSpeech Synthesis(Feature processing MFCC Log Area Ratio (LAR) Bark frequency scale FFT Power Spectrum LPC LSF F0 Likelihood Ratio Residual LP Mel Log Spectral Approximation (MLSA) MLSA filter Covariance Energy Delta, DeltaDelta =  Acoustic & Language model Viterbi decoding HMM (Baum-Welch) EM training WFST CART Harmonic plus Noise Modeling (HNM) Distance Euclidean Kullback-Leibler Mean Squared Log Spectral Distance (MS-LSD) Mahalanobis Itakura-Saito Symmetries Itakura RMS (root mean squared log spectral)[  -  = F, CNew Speech Functionality(Feature processing Bark scale Fundamental frequency Likelihood Ratio Covariance (Cholesky) MLSA CMS SB-MFCC WVF-MFCC Robust formant algorithm Split Levinson algorithm LPCC LFCC RMS energy Delta cepstral (Milner, High order) Pseudo LAR PLP. ?Acoustic & Language model HMM (Baum-Welch) HNM MLP WFST CART LDA, NLDA, QDA Minimum Bayes Risk (MBR) Maximum Likelihood Estimation NN (Elman predictive) Discrete Wigner Distribution Code Excited Linear Predictor Distance Kullback-Leibler Mean Squared Log Spectral Distance (MS-LSD) Itakura-Saito Symmetries Itakura RMSdPP PbP b,R SummaryIntel IPP 3.0 now covers most useful primitives for speech processing Speech enabled applications require still more primitives Developers and researches need more samples$ 1 Thank You !  H /"% & + , - .026?Pe ` G)ffft8>?" dZ@0x?lKZh@ i{ n?" dd@   @@``PX!$   @ I E` p>>8 0 me `(    6Dl "  MClick to edit Title Slide *  6n "P   RClick to edit Master text styles Second level Third level Fourth level Fifth level!     S  0s "`   `*a N BS   "BS   Z1?"lS WSeminar 2B  ND1?"  ZTy1?" XNNSU Lab 2    Z |1?" XSummer 2   Z41?" V2003 2B   ND1?"pB   ND1?"*b*B  ND1?"pbp2  T1?"En82  T1?"B  ND1?"uBB  ND1?";*;B  ND1?"*;*2  T1?";2  T1?"`2  T1?"EYn~  T1?"9`@  mSpeech processing" 2  0 "p   X*   3ffA8d޽h ?210218_MobilityDevCon_BAK_4 G)ffft8 $PCA_Intel_MobilityI 89 @(    <L #" `(  T Click to edit Master title style! !  < #" `    W#Click to edit Master subtitle style$ $  3ffA8d޽h ?210218_MobilityDevCon_BAK_4 G)ffft8, 0 pD|(  D D T|1 ?"P    X*   D T1 ?"    Z*  r D c $ ?"   D T1 ?" @  1@075F B5:AB0 B>@>9 C@>25=L "@5B89 C@>25=L '5B25@BK9 C@>25=L OBK9 C@>25=LM D Zx1 ?"`P   X*    D Z즩1 ?"`   Z*  H D 0޽h ? ̙33 <`( | < < TDP1 ?"P    V*    < TO1 ?"    X*    < ZA1 ?"`P   V*    < ZLE1 ?"`   X*   H < 0޽h ? ̙338 0$( )X    Tt*1?  tby Vitaly Horban" 2      T/1?9`@  mSpeech processing" 2r ! S 00   ( $ T 41? vgorban@unicyb.kiev.ua 2L  3ffA8d޽h ?210218_MobilityDevCon_BAK_4 f33f 89 L(    c $^"     c $X)"   B  s *޽h ? f33f* 8 X( X     s *I"       s *J"J0P  B  s *޽h ? 3ff3f6 8 Td(  T T  0x" @    T  0y"`0f  B T s *޽h ? 3ff3fk 8 $( " $ $ s *"PP    $ s *<"k@ P  9 $ 6p "u0 q  Feature processing LP Power Spectrum Cepstrum LSP Mel-scale values Mel-frequency filter bank Mel-cepstrum Linear scale values Acoustic & Language models Gaussian mixture Likelihood of an HMM state cluster HMM transition matrix0lKZl0KZ0lKZJ0KZlJhi{B $ s *޽h ? f33f 8 @(     0("PP'      0"k@ P    6é "`0@  `NFeature processing Power Spectral analysis (FFT) Linear predictive analysis (LPC) LP reflection coefficients LSP DCT RFC Cross correlation coefficients Covariance matrix Mel-scale cepstral analysis Derivative functions Energy normalization Acoustic & Language model VQ Weights, Means and Variances EM re-estimation Viterbi decoding0lKZ0KZ0lKZD0KZD( D hi{B  s *޽h ? f33f  89 &(     0"PP       0L"`@ P  P  6\ "`0@ 1  Feature processing Power Spectrum Reflection to Tilt, PitchmarkToF0, Unit Curve (RFC) LPC MEL LSF LP Reflection coefficients Energy normalization Acoustic & Language model Viterbi decoding 0lKZ0KZ0lKZ0KZ(  hi{B  s *޽h ? f33f  8 nf(     0T"PP      0"k@ P    6t "`0@  $Feature processing Derivative functions Spectrum Cepstrum Cross correlation Covariance matrix Energy normalization Filter bank Area Ratio Durbin s recursion Reflection coefficients (Schur) Gaussian probability Acoustic & Language model Viterbi decoding0lKZ0KZ0lKZ0KZB    hi{B  s *޽h ? f33f  8 ME (     00"PP   0    00"k@ P! 0 w  6 "`0@  Frequency Scale Conversion Mel scale Linear scale Transforms DFT FFT DCT Distance Euclidean Mahalanobis DTW (observation and reference vector sequences) Bhattacharya0lKP0KP 0lKP 0KP 0lKPT0KP   T   <  hi{B  s *޽h ? f33fl  8 0(     00"PP  0    00"k@ P 0 .  6 0 "`0@  Feature processing LPC Area ratio Spectrum Cepstrum RFC DCT LSP LSF Reflection coefficients RFC Autocorrelation coefficients Cross correlation coefficients Covariance matrix Mel-scale cepstral analysis Derivative functions Energy normalizationp0lKZ0KZ\ hi{B  s *޽h ? f33f  89 @(     0?0"PI  0    0@0"k@ P 0 D  6 "`0@  Speech Recognition Feature processing Model Evaluation Model Estimation Model Adaptation Vector Quantization Speech coding (ITU G.711, G.723.1, G.729) Linear PCM A-law Mu-law VQ given codebook0lKZZ0KZ*0lKZ*0KZZ**&   hi{B  s *޽h ? f33fk 8 P(     0Q0"PI  0    0R0"k@ PN 0 -  6% "`0@  Feature processing LPC Area ratio Spectrum Cepstrum RFC DCT LSP LSF Reflection coefficients Autocorrelation coefficients Cross correlation coefficients Covariance matrix Mel-scale cepstral analysis Derivative functions Energy normalization VQp0lKZ0KZX hi{B  s *޽h ? f33f6 8 `d(     0a0"PP  0    0b0"k@ P 0   6j0 "`0@ u  Model adaptation EM training algorithm Acoustic & Language model Viterbi decoding Likelihood of an HMM state cluster HMM transition matrix Speech coding (ITU G.711, G.723.1, G.729) Linear PCM A-law Mu-law VQ given codebook0lKZ0KZ0lKZJ0KZ*0lKZ*0KZI**  V  8<   hi{B  s *޽h ? f33f 8 OGp0(  0 0  0v0"PG  0  0  0v0"k@ P! 0 y 0 6D}0 "`0@  Feature processing PLP: filter, cepstrum, filter bank Relative spectra filtering of log domain coefficients (RASTA) Fundamental frequency (pitch) RMS energy Covariance (Cholesky) Energy (Log, dB, RMS, Power) LAR (Kelly-Lochbaum) Autocorrelation (Leroux-Guegen) Lattice (Burg) Equivalent Rectangular Bandwidths (ERB) Unit-triangular matrix (AR coef.) Expand formant bandwidths (LP) Third differential Hartley transform Diagonalisation of two Hermitian matrices (LDA, IMELDA)n0lKZ0KZ=0KP|      T   hi{B 0 s *޽h ? f33f 89  8(  8x 8 c $40P  0 x 8 c $ܐ0B@ a 0 x 8 c $0Or  0 H 8 0޽h ? G)ffft8 8 80 \(  \~ \ s *P   ~ \ s *h`@ @    \  0Ӡ mr   H \ 0޽h ? G)ffft8 8 80`(  `~ ` s *0P  0 ~ ` s *̙0`@  0  `  00 mr  0 H ` 0޽h ? G)ffft8 8 80d(  d~ d s *0P  0 ~ d s *0`@  0  d  0d0 mr  0 H d 0޽h ? G)ffft8 8 80h(  h~ h s *0P  0 ~ h s *0`@  0  h  0P0 mr Q 0 H h 0޽h ? G)ffft8 8 ,L( " , , c $$0"  0  , c $0"G  0 B , s *޽h ? f33f 8 ZR (       0("H      T1?  i Vitaly Horban" 2    )   T%1? vgorban@unicyb.kiev.ua  2LB   s *޽h ? f33f 0 `X (  XX X C D    X S (D @   "H X 0޽h ? ̙33 0 \ (  \X \ C D   0 \ S ,0D @  0 "H \ 0޽h ? ̙33 0 d (  dX d C D   0 d S 0D @  0 "H d 0޽h ? ̙33 0 l (  lX l C D   0 l S U0D @  0 "H l 0޽h ? ̙33 0 t (  tX t C D   2 t S $2D @  2 "H t 0޽h ? ̙33 0 ,(  ^  S D     c $D @   "H  0޽h ? ̙33  0  ,(  ^  S D     c $ĕD @   "H  0޽h ? ̙33  0 0,(  ^  S D     c $<D @   "H  0޽h ? ̙33  0 @,(  ^  S D     c $ǠD @   "H  0޽h ? ̙33  0 P,(  ^  S D   2  c $H2D @  2 "H  0޽h ? ̙33  0 ` ,(   ^   S D   2   c $t 2D @  2 "H   0޽h ? ̙33 0 p,(  ^  S D   2  c $2D @  2 "H  0޽h ? ̙33 0 ,(  ^  S D   2  c $2D @  2 "H  0޽h ? ̙33 0 $,(  $^ $ S D    $ c $ .D @   "H $ 0޽h ? ̙33 0 4,(  4^ 4 S D   2 4 c $2D @  2 "H 4 0޽h ? ̙33 0 X,(  X^ X S D   0 X c $0D @  0 "H X 0޽h ? ̙33r nt&v z  bi jB@Jj_ #TȗRrŠV:50«>`XxRJDJ(Oh+'0 px   D P \ ht| PowerPointGorbanVSC:\IntelSumSch2003\Workspace\Documents\Common\Different\Ppt\PCA_Intel_Mobility.pptGorbanV189Microsoft PowerPoint 7.0ace@P:q1@@w]KU@9/fhGg  E& &&#TNPP2OMi & TNPP &&TNPP    &&--2f- $ $ $** $**77 $77EE $EESS1f- $SSaa $aaoo $oo}}0f- $}} $ $/f- $.f- $ $-f- $ $,f- $+f- $*f- $)f- $##(f- $##11'f- $11>>&f- $>>LL%f- $LLZZ$f- $ZZhh#f- $hhvv"f- $vv f- $f- $f- $f- $f- $f- $f- $f- $f- $f- $f- $f- $**f- $**88f- $88FFf- $FFSS f- $SSaa f- $aaoo f- $oo}} f- $}}f- $f- $f- $f- $f- $---&&--Y(--qw@ LSwUSw0- @Times New RomanLSwUSw0- . 2 I2by  . .2 Ij Vitaly Horban   !.--1 -- @Times New RomanLSwUSw0- .!2 %Speech processing   .-- y~H-- @Times New RomanLSwUSw0- ).2 Intel( .@"Arialw@ LSwUSw0- . 2 h/. ). 2 f/.@Times New RomanLSwUSw0- ).-2 Integrated L( $ $ (. ).(2 rPerformance Primitives+ $ <$( + <$ .@Times New RomanLSwUSw0- ).E2 ?)vs. . ).02 |^Speech Libraries & Toolkits$$$+$ 5+ ". ).'2 Math Inside & Outside= $$$52$$.--p-- )@Times New RomanLSwUSw0- .2 (vgorban. . 2 @. .2 unicyb . . 2 .. . 2 kiev . . 2 T.. . 2 \ua.&c--d=(-- @Times New RomanLSwUSw0- .2 X1Seminar  .&P%S--%Q#Q--&--(--  . 2*t\)  6     #& // !  "#%(    O$$$R$ٶa|ջ.$c $0e0e     A@ AԔ Ԕ8c9     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S"@f3@8  H ʚ;2Nʚ;g4PdPd0pppp@ <4!d!d 0,(5<4dddd 0,(5<4BdBd 0,(5g4=d=d0p^p@ pp?* Summer 2003O*= Summer 2003 =Intel Integrated Performance Primitives vs. Speech Libraries & Toolkits Math Inside & Outside@P6 06cAgenda Comparison Intel IPP 3.0 and speech libraries & toolkits Overview mathematical methods for speech processing General assessment of Intel IPP 3.0 Summary6 x  Acronyms  ,LP Linear Prediction RELP Residual Linear Prediction PLP Perceptual Linear Prediction AR Area Ratios or Autoregressive LSP Line Spectrum Pairs LSF Line Spectral Frequencies MFCC Mel-Frequency Cepstrum Coefficients MLSA Mel Log Spectral Approximation DCT Discrete Cosine Transform DTW Dynamic Time Warping SVD Single Value Decomposition VQ Vector Quantization RFC Rise/Fall/Connections HMM Hidden Markov Model ANN Artificial Neural Network EM Expectation/Maximization >Acronyms (continue)&, $, CMS Cepstral Mean Subtraction MLP Multi Layer Perception LDA Linear Discriminant Analysis QDA Quadratic Discriminant Analysis NLDA Non-Linear Discriminant Analysis SVM Support Vector Machine DWT Discrete Wavelet Transformation LAR Log Area Ratio PLAR Pseudo Log Area Ratio GMM Gaussian Mixture Model WFST Weighted Finite State Transducer CART Classification and Regression Trees HNM Harmonic plus Noise Modeling MBR Minimum Bayes Risk SR Speech Recognition TTS Text-To-Speech synthesisVd$ !  5  IPP vs. CMU Sphinx,LFeature processing LP Spectrum Cepstrum MEL: filter, cepstrum, filter bank PLP: filter, cepstrum, filter bank Language model Context-free grammar N-gram model Acoustic model based on HMM Each HMM state  set of Gaussian mixture HMM order HMM position HMM transition matrix Baum-Welch training\"j9#")          !IPP vs. CSLU Toolkit,wFeature processing Power spectral analysis (FFT) Linear predictive analysis (LPC) PLP Mel-scale cepstral analysis (MEL) Relative spectra filtering of log domain coefficients (RASTA) First order derivative (DELTA) Energy normalization Language model Word pronunciation Lexical trees Grammars Acoustic model based on HMM/ANN VQ initialisation EM training Viterbi decoding* /A%47* /  ( Y/ # IPP vs. Festival,Feature processing Power spectrum Tilt to RFC, RFC to Tilt, RFC to F0 LPC MEL LSF LP Reflection coefficients Fundamental frequency (pitch) Root mean square energy Language model N-gram model Context-free grammar Regular expressions CART trees WFST Acoustic model Viterbi decodingFZ6E  ( d $  IPP vs. ISIP , |Feature processing Derivative functions Spectrum Cepstrum Cross correlation Covariance matrix Covariance (Cholesky) Energy (Log, dB, RMS, Power) Filter bank Log Area Ratio (Kelly-Lochbaum) Autocorrelation (Durbin recursion, Leroux-Guegen) Lattice (Burg) Reflection coefficients Gaussian probability Acoustic & Language model (HMM) N-gram model Viterbi decoding Baum-Welch trainingPP P2PK3  # -  x  %    > Q ' IPP vs. MATLAB,Frequency Scale Conversion Mel   !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~      !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~Root EntrydO)PANuPictures.$Current UserPSummaryInformation( PowerPoint Document(ADocumentSummaryInformation8L / 0DTimes New Roman<4|d0|Wo 0DArialNew Roman<4|d0|Wo 0h" DWingdingsRoman<4|d0|Wo 0hc  .  @n?" dd@  @@`` 2*t\)  6     #& // !  "#%(    O$$$R$ٶa|ջ.$c $0e0e     A@ AԔ Ԕ8c9     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S"@f3@8  H ʚ;2Nʚ;g4PdPd0pppp@ <4!d!d 0,(5<4dddd 0,(5<4BdBd 0,(5g4=d=d0p^p@ pp?* Summer 2003O*= Summer 2003 =Intel Integrated Performance Primitives vs. Speech Libraries & Toolkits Math Inside & Outside@P6 06cAgenda Comparison Intel IPP 3.0 and speech libraries & toolkits Overview mathematical methods for speech processing General assessment of Intel IPP 3.0 Summary6 x  Acronyms  ,LP Linear Prediction RELP Residual Linear Prediction PLP Perceptual Linear Prediction AR Area Ratios or Autoregressive LSP Line Spectrum Pairs LSF Line Spectral Frequencies MFCC Mel-Frequency Cepstrum Coefficients MLSA Mel Log Spectral Approximation DCT Discrete Cosine Transform DTW Dynamic Time Warping SVD Single Value Decomposition VQ Vector Quantization RFC Rise/Fall/Connections HMM Hidden Markov Model ANN Artificial Neural Network EM Expectation/Maximization >Acronyms (continue)&, $, CMS Cepstral Mean Subtraction MLP Multi Layer Perception LDA Linear Discriminant Analysis QDA Quadratic Discriminant Analysis NLDA Non-Linear Discriminant Analysis SVM Support Vector Machine DWT Discrete Wavelet Transformation LAR Log Area Ratio PLAR Pseudo Log Area Ratio GMM Gaussian Mixture Model WFST Weighted Finite State Transducer CART Classification and Regression Trees HNM Harmonic plus Noise Modeling MBR Minimum Bayes Risk SR Speech Recognition TTS Text-To-Speech synthesisVd$ !  5  IPP vs. CMU Sphinx,LFeature processing LP Spectrum Cepstrum MEL: filter, cepstrum, filter bank PLP: filter, cepstrum, filter bank Language model Context-free grammar N-gram model Acoustic model based on HMM Each HMM state  set of Gaussian mixture HMM order HMM position HMM transition matrix Baum-Welch training\"j9#")          !IPP vs. CSLU Toolkit,wFeature processing Power spectral analysis (FFT) Linear predictive analysis (LPC) PLP Mel-scale cepstral analysis (MEL) Relative spectra filtering of log domain coefficients (RASTA) First order derivative (DELTA) Energy normalization Language model Word pronunciation Lexical trees Grammars Acoustic model based on HMM/ANN VQ initialisation EM training Viterbi decoding* /A%47* /  ( Y/ # IPP vs. Festival,Feature processing Power spectrum Tilt to RFC, RFC to Tilt, RFC to F0 LPC MEL LSF LP Reflection coefficients Fundamental frequency (pitch) Root mean square energy Language model N-gram model Context-free grammar Regular expressions CART trees WFST Acoustic model Viterbi decodingFZ6E  ( d $  IPP vs. ISIP , |Feature processing Derivative functions Spectrum Cepstrum Cross correlation Covariance matrix Covariance (Cholesky) Energy (Log, dB, RMS, Power) Filter bank Log Area Ratio (Kelly-Lochbaum) Autocorrelation (Durbin recursion, Leroux-Guegen) Lattice (Burg) Reflection coefficients Gaussian probability Acoustic & Language model (HMM) N-gram model Viterbi decoding Baum-Welch trainingPP P2PK3  # -  x  %    > Q ' IPP vs. MATLAB,Frequency Scale Conversion Mel scale Equivalent rectangular Bandwidths (ERB) Transforms FFT (real data) DCT (real data) Hartley (real data) Diagonalisation of two Hermitian matrices (LDA, IMELDA) Vector distance Euclidean Squared Euclidean Mahalanobis Itakura (AR, Power spectra) Itakura-Saito (AR, Power spectra) COSH (AR, Power spectra) Speech enhancement Martin spectral subtraction algorithmP2P PlPPPP&P (  L  W  & W  ( IPP vs. MATLAB (continue), %LPC analysis and transforms Area ratios Autoregressive or AR Power spectrum Cepstrum DCT Impulse response (IR) LSP LSF Reflection coefficients Unit-triangular matrix containing the AR coefficients Autocorrelation coefficients Expand formant bandwidths of LPC filter Warp cepstral (Mel, Linear)f s6( & ) IPP vs. MATLAB (continue), $Speech synthesis Rosenberg glottal model Liljencrants-Fant glottal model Speech Recognition Mel-cepstrum Mel-filter bank Cepstral & variances to power domain Gaussian Mixture Speech coding (ITU G.711) Linear PCM A-law Mu-law VQ using K-means algorithm VQ using the Linde-Buzo-Gray algorithm9TZ7R  B  )       -        * IPP vs. HTK , Feature processing LPC Spectral coefficients Cepstral coefficients Reflection coefficients Gaussian distribution K-means procedure PLP Autocorrelation coefficients Covariance matrix Mel-scale filter bank MFCC Third differential Energy VQ codebookf_J /IPP vs. HTK (continue) ,  Model adaptation Maximum Likelihood Linear Regression (MLLR) EM technique Bayesian adaptation or Maximum Aposteriori Approach (MAP) Acoustic & Language model based on HMM Grammar N-gram model Viterbi training Baum-Welch training Speech coding Linear PCM A-law Mu-lawt':, ;'  B 3   5Possible extension IPP 3.0(Model adaptation Maximum Likelihood Linear Regression (MLLR) Bayesian adaptation or Maximum Aposteriori Approach (MAP) Model evaluation Itakura (AR, Power spectra) Itakura-Saito (AR, Power spectra) COSH (AR, Power spectra) Speech synthesis Rosenberg glottal model Liljencrants-Fant glottal model Speech enhancement Martin spectral subtraction Speech coding VQ using K-means algorithm VQ using the Linde-Buzo-Graym Acoustic model based on HMM Baum-Welch training4PgPPWPP8PPPP9PPPgW8    9f g )      / (        7Speaker Characteristics(JFeature processing Preemphasize Cepstral Energy Cepstral Mean Subtraction (CMS) MFCC, LPCC, LFCC LPC (to Cepstral, to LSF) Residual Prediction Mel-cepstral Fundamental frequency (F0) LSF (Bark scale) RMS energy Levinson-Durbin recursion Covariance (Cholesky) Delta cepstral (Milner, High order) Pseudo Log Area Ratio (PLAR) DWT VQ8     WJ Acoustic model Distance Bhattacharya Euclidean DTW Viterbi decoding EM (Lloyd) K-means (Lloyd) PLP MLP Twin-output MLP LDA NLDA Generative models GMM HMM (Baum-Welch) Q 1  @Speech Processing(Feature processing LPC LSP F0 Levinson-Durbin recursion Tilt Gaussian Acoustic & Language model Baum-Welch training Viterbi decoding CART Statistical language modeling3H(# |Speech enhancement Speech Analysis Discrete Wigner Distribution DWT Pitch Determination Code Excited Linear Predictor (CELP)N#Z#%|ASpeech Recognition(Feature processing DCT MFCC Mel-frequency log energy coefficients (MFLEC) Subband (SB-MFCC) CMS Within Vector Filtered (WVF-MFCC) Robust Formant (RF) algorithm Split Levinson Algorithm (SLA) Vector Quantization VQ correlation Single VQ Joint VQj"7",J% Acoustic & Language model Viterbi decoding LDA QDA MLP PLP EM re-estimation Minimum Bayes Risk (MBR) Maximum Likelihood Estimation (MLE) NN (Elman predictive) HMM (Baum-Welch) GMM Buried Markov Model Decision tree state clustering WFST Dynamic Bayesian NetworksdBSpeech Synthesis(Feature processing MFCC Log Area Ratio (LAR) Bark frequency scale FFT Power Spectrum LPC LSF F0 Likelihood Ratio Residual LP Mel Log Spectral Approximation (MLSA) MLSA filter Covariance Energy Delta, DeltaDelta =  Acoustic & Language model Viterbi decoding HMM (Baum-Welch) EM training WFST CART Harmonic plus Noise Modeling (HNM) Distance Euclidean Kullback-Leibler Mean Squared Log Spectral Distance (MS-LSD) Mahalanobis Itakura-Saito Symmetries Itakura RMS (root mean squared log spectral)[  -  = F, CNew Speech Functionality(Feature processing Bark scale Fundamental frequency Likelihood Ratio Covariance (Cholesky) MLSA CMS SB-MFCC WVF-MFCC Robust formant algorithm Split Levinson algorithm LPCC LFCC RMS energy Delta cepstral (Milner, High order) Pseudo LAR PLP. ?Acoustic & Language model HMM (Baum-Welch) HNM MLP WFST CART LDA, NLDA, QDA Minimum Bayes Risk (MBR) Maximum Likelihood Estimation NN (Elman predictive) Discrete Wigner Distribution Code Excited Linear Predictor Distance Kullback-Leibler Mean Squared Log Spectral Distance (MS-LSD) Itakura-Saito Symmetries Itakura RMSdPP PbP b,R SummaryIntel IPP 3.0 now covers most useful primitives for speech processing Speech enabled applications require still more primitives Developers and researches need more samples$ 1 Thank You !  H /"% & + , - .026?P8\ 05( )X    Tt*1?  tby Vitaly Horban" 2      T/1?9`@  mSpeech processing" 2r ! S 00   ( $ T 41? vgorban@unicyb.kiev.ua 2L T BS % "BS & TT1?"lS WSeminar 2B ' HD1?" ( T@1?" XNNSU Lab 2  ) T1?" XSummer 2 * T1?" V2003 2B + HD1?"pB , HD1?"*b*B - HD1?"pbp2 . N1?"En82 / N1?"B 0 HD1?"uBB 1 HD1?";*;B 2 HD1?"*;*2 3 N1?";2 4 N1?"`2 5 N1?"EYn~  3ffA8d޽h ?210218_MobilityDevCon_BAK_4 f33fl  8 0(     00"Pp  0    00"k@ P 0 .  6 0 "`0@  Feature processing LPC Area ratio Spectrum Cepstrum RFC DCT LSP LSF Reflection coefficients RFC Autocorrelation coefficients Cross correlation coefficients Covariance matrix Mel-scale cepstral analysis Derivative functions Energy normalizationp0lKZ0KZ\ hi{B  s *޽h ? f33f  8 @(     0?0"P w  0    0@0"k@ P 0 D  6 "`0@  Speech Recognition Feature processing Model Evaluation Model Estimation Model Adaptation Vector Quantization Speech coding (ITU G.711, G.723.1, G.729) Linear PCM A-law Mu-law VQ given codebook0lKZZ0KZ*0lKZ*0KZZ**&   hi{B  s *޽h ? f33f 8 OGp0(  0 0  0v0"P  0  0  0v0"k@ P! 0 y 0 6D}0 "`0@  Feature processing PLP: filter, cepstrum, filter bank Relative spectra filtering of log domain coefficients (RASTA) Fundamental frequency (pitch) RMS energy Covariance (Cholesky) Energy (Log, dB, RMS, Power) LAR (Kelly-Lochbaum) Autocorrelation (Leroux-Guegen) Lattice (Burg) Equivalent Rectangular Bandwidths (ERB) Unit-triangular matrix (AR coef.) Expand formant bandwidths (LP) Third differential Hartley transform Diagonalisation of two Hermitian matrices (LDA, IMELDA)n0lKZ0KZ=0KP|      T   hi{B 0 s *޽h ? f33fr$(( 9w=5BGDJ( / 0DTimes New Roman<4|d0|Wo 0DArialNew Roman<4|d0|Wo 0h" DWingdingsRoman<4|d0|Wo 0hc  .  @n?" dd@  @@``2 NNSU Lab  .--GY--  .2 ;Summer  .--GM--  . 2 ;W2003 .&S--%Q--&&14--%22--&&--%--&G--5.--G--T)M!--&PS--%QQ--&&1S--%Q2--&&14--%22--&G--6.--G--UM--G----&--"System 0-&TNPP & ՜.+,0    oNNSUoA2 Times New RomanArial WingdingsPCA_Intel_MobilityIntel Integrated Performance Primitives vs. Speech Libraries & Toolkits Math Inside & OutsideAgenda Acronyms Acronyms (continue)IPP vs. CMU SphinxIPP vs. CSLU ToolkitIPP vs. Festival IPP vs. ISIPIPP vs. MATLABIPP vs. MATLAB (continue)IPP vs. MATLAB (continue) IPP vs. HTKIPP vs. HTK (continue)Possible extension IPP 3.0Speaker CharacteristicsSpeech ProcessingSpeech RecognitionSpeech SynthesisNew Speech FunctionalitySummary Thank You !    (_Victor P. GergelVictor P. Gergelscale Equivalent rectangular Bandwidths (ERB) Transforms FFT (real data) DCT (real data) Hartley (real data) Diagonalisation of two Hermitian matrices (LDA, IMELDA) Vector distance Euclidean Squared Euclidean Mahalanobis Itakura (AR, Power spectra) Itakura-Saito (AR, Power spectra) COSH (AR, Power spectra) Speech enhancement Martin spectral subtraction algorithmP2P PlPPPP&P (  L  W  & W  ( IPP vs. MATLAB (continue), %LPC analysis and transforms Area ratios Autoregressive or AR Power spectrum Cepstrum DCT Impulse response (IR) LSP LSF Reflection coefficients Unit-triangular matrix containing the AR coefficients Autocorrelation coefficients Expand formant bandwidths of LPC filter Warp cepstral (Mel, Linear)f s6( & ) IPP vs. MATLAB (continue), $Speech synthesis Rosenberg glottal model Liljencrants-Fant glottal model Speech Recognition Mel-cepstrum Mel-filter bank Cepstral & variances to power domain Gaussian Mixture Speech coding (ITU G.711) Linear PCM A-law Mu-law VQ using K-means algorithm VQ using the Linde-Buzo-Gray algorithm9TZ7R  B  )       -        * IPP vs. HTK , Feature processing LPC Spectral coefficients Cepstral coefficients Reflection coefficients Gaussian distribution K-means procedure PLP Autocorrelation coefficients Covariance matrix Mel-scale filter bank MFCC Third differential Energy VQ codebookf_J /IPP vs. HTK (continue) ,  Model adaptation Maximum Likelihood Linear Regression (MLLR) EM technique Bayesian adaptation or Maximum Aposteriori Approach (MAP) Acoustic & Language model based on HMM Grammar N-gram model Viterbi training Baum-Welch training Speech coding Linear PCM A-law Mu-lawt':, ;'  B 3   5Possible extension IPP 3.0(Model adaptation Maximum Likelihood Linear Regression (MLLR) Bayesian adaptation or Maximum Aposteriori Approach (MAP) Model evaluation Itakura (AR, Power spectra) Itakura-Saito (AR, Power spectra) COSH (AR, Power spectra) Speech synthesis Rosenberg glottal model Liljencrants-Fant glottal model Speech enhancement Martin spectral subtraction Speech coding VQ using K-means algorithm VQ using the Linde-Buzo-Graym Acoustic model based on HMM Baum-Welch training4PgPPWPP8PPPP9PPPgW8    9f g )      / (        7Speaker Characteristics(JFeature processing Preemphasize Cepstral Energy Cepstral Mean Subtraction (CMS) MFCC, LPCC, LFCC LPC (to Cepstral, to LSF) Residual Prediction Mel-cepstral Fundamental frequency (F0) LSF (Bark scale) RMS energy Levinson-Durbin recursion Covariance (Cholesky) Delta cepstral (Milner, High order) Pseudo Log Area Ratio (PLAR) DWT VQ8     WJ Acoustic model Distance Bhattacharya Euclidean DTW Viterbi decoding EM (Lloyd) K-means (Lloyd) PLP MLP Twin-output MLP LDA NLDA Generative models GMM HMM (Baum-Welch) Q 1  @Speech Processing(Feature processing LPC LSP F0 Levinson-Durbin recursion Tilt Gaussian Acoustic & Language model Baum-Welch training Viterbi decoding CART Statistical language modeling3H(# |Speech enhancement Speech Analysis Discrete Wigner Distribution DWT Pitch Determination Code Excited Linear Predictor (CELP)N#Z#%|ASpeech Recognition(Feature processing DCT MFCC Mel-frequency log energy coefficients (MFLEC) Subband (SB-MFCC) CMS Within Vector Filtered (WVF-MFCC) Robust Formant (RF) algorithm Split Levinson Algorithm (SLA) Vector Quantization VQ correlation Single VQ Joint VQj"7",J% Acoustic & Language model Viterbi decoding LDA QDA MLP PLP EM re-estimation Minimum Bayes Risk (MBR) Maximum Likelihood Estimation (MLE) NN (Elman predictive) HMM (Baum-Welch) GMM Buried Markov Model Decision tree state clustering WFST Dynamic Bayesian NetworksdBSpeech Synthesis(Feature processing MFCC Log Area Ratio (LAR) Bark frequency scale FFT Power Spectrum LPC LSF F0 Likelihood Ratio Residual LP Mel Log Spectral Approximation (MLSA) MLSA filter Covariance Energy Delta, DeltaDelta =  Acoustic & Language model Viterbi decoding HMM (Baum-Welch) EM training WFST CART Harmonic plus Noise Modeling (HNM) Distance Euclidean Kullback-Leibler Mean Squared Log Spectral Distance (MS-LSD) Mahalanobis Itakura-Saito Symmetries Itakura RMS (root mean squared log spectral)[  -  = F, CNew Speech Functionality(Feature processing Bark scale Fundamental frequency Likelihood Ratio Covariance (Cholesky) MLSA CMS SB-MFCC WVF-MFCC Robust formant algorithm Split Levinson algorithm LPCC LFCC RMS energy Delta cepstral (Milner, High order) Pseudo LAR PLP. ?Acoustic & Language model HMM (Baum-Welch) HNM MLP WFST CART LDA, NLDA, QDA Minimum Bayes Risk (MBR) Maximum Likelihood Estimation NN (Elman predictive) Discrete Wigner Distribution Code Excited Linear Predictor Distance Kullback-Leibler Mean Squared Log Spectral Distance (MS-LSD) Itakura-Saito Symmetries Itakura RMSdPP PbP b,R SummaryIntel IPP 3.0 now covers most useful primitives for speech processing Speech enabled applications require still more primitives Developers and researches need more samples$ 1 Thank You !  H /"% & + , - .026?PrHG D