ࡱ; vtrp  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuwxyz{|}~R FU~;CompObj\WordDocument HObjectPoolmQ~;U~;  FMicrosoft Word 6.0 DocumentNB6WWord.Document.6; L%#TPࡱ;  FMicrosoft Word 6.0 PictureNB6WWord.Picture.6ࡱ; ࡱ;  Oh+'0bܥhS e%1H%.DrDrrDrDrDrDrDEEEEEE,E(EGLTEjEjEjEjEjEjEjEG:GGGGGG&0HXH?GrDjEjEjEjEjEGjErDrDjETEjEjEjEjErDjErDjEGD8D^rDrDrDrDjEGjEjEMEG Studies of Vowel Processing in Auditory Cortex Colin Phillips U. of DelawareKrishna Govindarajan Alec Marantz M.I.T.David Poeppel Tim Roberts Howard Rowley U.C. San FranciscoElron Yellin Mark of the Unicorn Inc. 1. Cues to Speaker and Category in Vowels Vowels are acoustically complex sounds, consisting of bands of energy at a fundamental frequency (F0) and a series of prominent harmonics of the fundamental frequency (formants). The fundamental frequency carries information about prosody and speaker voice. The formants carry information about the phonetic category of the vowel.  Figure 1: Spectrogram of the vowel /u/ taken from the word boot For example, the first formant (F1) carries information about the height of the vowel: high vowels (e.g. /i/, /u/) have a low F1 (~300Hz); low vowels (e.g. /a/, //) have a high F1 (~700Hz). The second formant carries information about the frontness/backness of the vowel: front vowels (e.g. /i/, /e/) have high F2 values (~2000Hz), and back vowels (e.g. /u/, /a/) have lower F2 values (~1000Hz). The studies described here demonstrate an evoked response in auditory cortex which appears to track the formant structure of vowels rather than the fundamental frequency. 2. Pure tones: pitch affects M100 latency We focus in this poster on analyses of the auditory M100 response, a well-known evoked response generated in supratemporal auditory cortex approx. 100ms after the onset of any acoustic stimulus. Figure 2a illustrates the characteristic M100 waveform, recorded with a 37-channel MEG array, and the stages involved in obtaining a source-localization from the magnetic waveforms (the whole-head reconstruction in Figure 2d shows somatosensory rather than auditory responses).  Figure 2: Measurement and localization of evoked brain responses using Magnetic Source Imaging The latency of evoked M100 responses varies systematically as a function of the pitch of tone stimuli (Roberts & Poeppel 1996; Ragot & Lepaul-Ercole 1996). Frequencies at 1000Hz or above evoke the shortest latency M100s; lower frequencies evoke progressively longer latency M100s.  EMBED Word.Picture.6  Figure 3: Effect of tone-frequency on M100 latency (n=5) The latency of M100 in response to a tone can be predicted by the following empirical formula: L = 3/f + k where f is the tone frequency, and k is a time constant that varies across subjects from 90-120 ms. This formula can be used to predict the M100 latency difference for two tones of different frequencies. We use this finding as a tool for understanding the cause of the systematic M100 latency changes in response to different stimuli in our vowel experiments. 3. Three-formant vowels To investigate responses to vowels as a function of pitch (F0), phonetic category (formants) and hemisphere, we presented 3 synthetic vowels (/a/, /i/ and /u/) at two fundamental frequencies (100Hz, male voice (m) and 200Hz, female voice (f)) to 6 subjects while recording the evoked neuromagnetic field from the left or right temporal cortices using a 37-channel biomagnetometer. The formant values are shown in Table 1. Table 1: Vowel formantsspectral composition (in Hz) StimulusF0F1bandwidthF2bandwidthF3bandwidtha(m)100710901100100254090a(f)200850901220100281090i(m)100280902250100289090i(f)200310902790100331090u(m)10031090870100225090u(f)20037090950100267090 Procedure Vowel stimuli were synthesized using a Klatt formant synthesizer (SenSyn, Sensimetrics Inc.). Each vowel was presented 100 times in pseudorandom order to the ear contralateral to the MEG dewar. ISIs varied randomly from 7501250ms. We measured the extracranial magnetic field using a 37-channel first-order gradiometer (Magnes, BTi, San Diego, CA). Recordings took place in a shielded room. Subjects were lying down, and the dewar containing the sensor array was positioned to record from auditory cortex. The appropriate position was determined by first localizing the generator of the M100 response to series of 1000Hz and 200Hz tones. Responses were recorded with a sampling rate of 1kHz and a bandwidth of 400Hz. Responses were then averaged according to event type and filtered using a low-pass filter from 1-20Hz. For each of the averaged and filtered responses a single-dipole source localization model was computed using an interative least-squares algorithm. M100 latencies were identified as the time interval between 80 and 150ms post-stimulus onset with the largest root mean square (RMS) of the field strength across the 37 channels. Results Measured M100 latencies varied systematically as a function of vowel category differences (F(2,5)=27.2, p<0.0001), but there were no significant changes due to F0 differences (figure 4). This result was significant in both hemispheres.  Figure 4: Latency variation due to category (n=6) This result is consistent with Roberts & Poeppels findings about effects of tone-frequency on M100 latency. The fastest responses were to vowels with the most energy around 1000Hz. More specifically, M100 latencies tracked variation in the first formant (F1) of the vowels extremely closely (r = 0.97). The consistently earlier M100 to female vowels may be due to the higher F1 values in the female vowels. However, this could also be an effect of the higher F0 in the female vowels. The second experiment addresses this uncertainty. There was no significant effect of F0 or formant values on the localization of M100 responses; responses to all vowels localized reliably to the same region of supratemporal auditory cortex (figures 56).  Figure 5: Localizations of M100 responses to 3-formant /a/s in left and right auditory cortices  Figure 6: Source localizations of M100 response to vowels (same subject as Figure 5) The results of this experiment indicate that latency differences in M100 responses to vowels are determined by the energy bands that carry category information (formants) rather than the energy band that carries prosody and voice information (F0). 4. Single-formant vowels The first vowel experiment was intended to determine the relative contribution of the fundamental frequency and the formants to early responses to vowel sounds. The results suggested that the value of the first formant, which carries information about vowel height, is the main factor determining M100 latencies. In a subsequent experiment we set out to test this correlation further, by isolating the contribution of F1.  Figure 7: Waveforms of single-formant vowel stimuli Although naturally produced vowels contain at least 3 strong formants, it is possible to synthetically generate vowels with just a single formant in addition to the fundamental frequency. In the case of the vowels /a/ and /u/ the sounds sound somewhat unnatural but are quite easily identified by subjects. For this study we used single-formant versions of /a/ and /u/ at each of two fundamental frequencies. In this experiment we did not adjust the value of F1 in the female versions to accommodate the higher fundamental frequency. Thus F0 and F1 varied entirely independently.  Figure 8: Effect of tone frequency on M100 latency replicated for subjects in this experiment (n=9) MEG recordings were made from left temporoparietal cortex of 9 subjects with normal hearing while 100 tokens of each of the four single-formant vowels was presented to the contralateral ear. Pretests using pure tones on each of the subjects in this study replicated the effect of frequency on M100 latency, showing almost identical effect size and means to Roberts & Poeppel (1996). The different predictions, according to whether F0 or F1 determines M100 latency, are shown in Figures 910. If latency variation is due to F0 then female vowels are predicted to show approx. 10ms faster M100s; if latency variation in vowels is due to F1 then /a/ is predicted to show approx. 5ms faster M100s. If both F0 and F1 affect response latencies, then we expect responses to /a/ to be faster than responses to /u/, with responses to female vowels additionally being faster for both categories.   Figure 9: Prediction if F0 determines M100 latencyFigure 10: Prediction if F1 determines M100 latency The results showed a significantly shorter latency M100 in response to single formant /a/ (F(1,8)=4.2, p < 0.05), with an effect size of 5.2ms. This result strongly supports the claim that M100 varies as a function of the F1 value of a vowel.  Figure 11: F1 determines M100 latency (n=9) There was no evidence for an effect of F0 on M100 latencies, and no effect of either category or voice on localizations of these evoked responses. 5. Why formants determine M100 variation Having established that M100 latencies vary as a function of the formants of vowels, and are not affected by variations in F0, we now need to ask why this is the case. Two possible explanations are: Speech specific account: formant values are the major determinant of response latencies because in speech sounds they contain the information that enables speakers to identify phoneme categories. General auditory account: the auditory system is more sensitive to energy in the frequency range where formants occur, regardless of whether the energy is part of a speech sound or not. Energy in F0 is just not so salient to the auditory system. Although we know of no direct support for the speech-specific account, an ERP study by Ragot & Lepaul Ercole (1996) does challenge our conclusion that the formant frequencies rather than F0 determine the latency of the M100. They independently varied the fundamental and spectrum of tone complexes and found N100 latency variation due to changes in the fundamental rather than to changes in the spectrum. Several properties of their stimuli make their results difficult to compare with ours. Most crucially, the energy in the Ragot & Lepaul-Ercole sounds was heavily concentrated in the F2 range, rather than the F1 range. However, results of a pilot MEG study with tone complexes in our lab are more consistent with the general auditory account. We created complexes of pairs of pure tones, with frequencies and amplitudes matching those of the stimuli in the single-formant vowels experiment. In one subject tested, the tone complexes did not sound remotely speechlike, but variation in the pseudo-F1 caused the greatest changes in M100 latency.  Figure 12: Pilot comparison of single-formant vowels and matching tone complexes (n=1) 6. Conclusion Vowel sounds contain energy at a number of different frequencies, which between them convey both speaker and category information. Our results show that the auditory M100 response varies as a function of category-relevant information (formants) rather than voice-relevant information (fundamental frequency); specifically, the latency of M100 closely tracks the value of the first formant of a vowel, which contains information about the height of the vowel. Although we have shown that M100 latency correlates closely with linguistically relevant properties of vowels, it remains for future work to determine whether this response pattern is particular to speech sounds, or whether it reflects general properties of auditory processing. Acknowledgment We would like to acknowledge the support of the following organizations for this research: NEC Corporation, NSF SGER grant #SBR-9423242, the McDonnell-Pew Center for Cognitive Neuroscience at MIT, the NSF-funded RTG Program Language: Acquisition and Computation (#DIR9113607), NIH grant DC00194-02, and the JST funded project 'Mind Articulation' (co-PIs W. O'Neil, MIT and Y. Miyashita, U. of Tokyo). =|HH(FG(HH(d'@Y: I h) S&WordMicrosoft Word&Word   2 @@$Use Word 6.0c or later to 2 @$view Macintosh picture.      0I:9"y #K%%DSIDICT:_cv userdict /_cv known not currentdict /bu known and{bu}{save}ifelse userdict /_cv known not dup {userdict /_cv 20 dict put}if _cv begin /bdf{bind def}bind def /isDeviceColor {/currentcolorspace where{pop currentcolorspace 0 get dup /DeviceGray eq exch dup /DeviceRGB eq exch /DeviceCMYK eq or or}{T}ifelse}bdf /setcmykcolor where{/setcmykcolor get /cvcmyk exch def}{/cvcmyk{1 sub 4 1 roll 3{3 index add neg dup 0 lt{pop 0}if 3 1 roll}repeat setrgbcolor pop}bdf }ifelse /stg{isDeviceColor {cf ca /cs load setscreen setgray}{pop}ifelse}bdf /strgb{isDeviceColor {cf ca /cs load setscreen setrgbcolor}{pop pop pop}ifelse}bdf /stcmyk{isDeviceColor {cf ca /cs load setscreen cvcmyk}{pop pop pop pop}ifelse}bdf /min1{dup 0 eq{pop 1}if}bdf currentscreen/cs exch def/ca exch def/cf exch def end currentdict /bn known and{bn}{restore}ifelse ##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #d0drw27 } ##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #HH` ff33̙ ff 33 ff33ffffffffffff33ff33333333ff333333 !ff"33#$%&'ff(33)*+,̙-ff.33/0̙1̙2̙3̙ff4̙335̙6ff7ff8ff9ffff:ff33;ff<33=33>33?33ff@3333A33BCDEffF33GHIJKffL33MNOP̙QffR33STUVWffX33YZff[ff\ff]ffff^ff33_ff`33a33b33c33ffd3333e33fghiffj33klffmffnffoffffpff33qffrffsfftff̙uffffvff33wffxffyffzff{ffff|ff33}ff~ffffffffffffffffffffff33ffffff33ff33ff33ff33ffff3333ff33ffffffffffff33ff33333333ff333333333333̙33ff33333333333333ff33333333ff33ff33ff33ffff33ff3333ff3333333333333333ff333333333333333333ff333333ff33̙ff33ff33ffffffffffff33ff33333333ff333333ff33ٻڪۈwwUUDD""wwUUDD""wwUUDD""wwwwwwUUUUUUDDDDDD""""""$    !!"',( 8+L+U++N++R++U+^+++j+ +++++++++++++z+++++++ +++++ ++++ +++++++++++++++++++ ++ ++++++++++++++ +++++++VV++V++ ++++++++++++++VVVVVVVVV++++V+++++++++++++VV++VV+++++ +VV++++V+*++++VV++V+++VV+++++++++ VV+#++++VVVV+++++++VVV++++V+++VV++++++ +V V++++/VVV+++++VV+++++++++ ++VVVVV+++++9VVV++++++V++++++++V+VV ++V++?++VV++++++V++++++V+V+VVVVV++V+++++++.++VV++++++V++ +V++V V +++++VV+VV+ ++VV+++VV+V+V V++++VV+V+++V++V+++VV+VV VV++++V++V++VV+V+++V+V V++VVV++++VV+++VV++V++*+VV+VVVV*+V++++VV+!+V++V++*+VV+VV.++++VVV+ +V++VV++V++2+VV+V++VV+)+V++VCVV++V+++V+VV+)+V++V+CV+VVV+++++VV+VV+VV+)+V++VVVV+,VV+V+V+VV+++V++V+ ++)+++VV+VV++VV+++V+V+VV+++++VIV+V++VV++++V+VV+++)V++++V++++V5V+V+V+++V+++++ +V++V+VQV+V+VV++V+V++V++V+++ VVVV+VV+VV++V+VV++++V++V++V+V+V+VV+V+V++*V+V+++VV+ ++ VVVDVVV++VV+V+++V+V+VVVV"V++VV+V+V+++ 2VVVVVV++VVVV++++V+VVVV++VVVV+++++V++V/VVV+VVVV+++++++ VVVV+V+V+VVVVVV++++V+V+(V+++V++VVVV++++ VV+V+VV++V+ ++++V+++V+V++V (V++V+V++++ ++++++VV+V+ V+V++V+++++++++++VV++VV+ +VVVVV++++VV+++++VV+++ +++++VVVV++VV+++ +VV+V++++V+++ +++++V+++++++V++VVVVV+V+++++++++++++VV++V++++9VV+++VV+V++++ ++++V++V+++++++VV++VV+++++++V++VV+++6++++V+VV+V++V++++ +V+++++,+VV+V+V+++++++++++(+VVV++VVV+++V++++z+V+?++V++V+V++V+++V+ +V-VV++V++++|+V++V-VV++VV+V++z+V++V-VV+VV+V+++t+V5+VV+V++VV+V+++e+V++V+V+V+V+ +++k+V+V++VV+V++ ++++m+V+V++VVV+V++++++`+V++VVV++V++++++a+V+8+VV+V++++Z+V+ +V+V++`+V+V+++++N+V+++VV++ ++++W+VV+++V+++++Z +VV+ ++++++V++++++p +V++++++ +V++++++e +V+ +++++V++++++++j++++(+V++++VV++++j++ /+++++V++++VV+ ++f ++ ++ V+V VV +++h+2++++V+++V+V+++++w+++V++VV+VVV+V++ +{ V+++V+V+ VV++V +++V++VV++V +VV+++#++++++V+++++V+++++VV+V+VVVVV++++ V++0+++++V++VVV+VV++++++++ V+++++ V+V+V+V++++ +VVV V+VV+VVV+VV++++++++ +VV+VVVVVV+++++++++ +VV+V +V++++++ V+VV+ VVV+++++++++z V++V+VV+++++++}V V+++ VVV++++++V+ +V+VV++++++++#++VVV+ VV+++++ +V++VV V++ +++++ +VVVV+VV+VVVVV++ +++ +VV+VV+VVVV++++ +++V+VV'VV++VV+V++++++++++V+0+VV+V++++++++V++@+V+VVVVV+VV++++++ +VV++++++V+V++V+V++++++++++V+++V V+V+VV++++~+ ++1+V++++VV++V+++++++{ +++++ +V*V+++VV+VV+++p++++>+++++VV+++++o+++8+++++V+VV+++ j+++++++V+V+V+++++++l++++#++VV+VVV++++++m++ +'+++VV+VV+V++p++++++V+V+V+++ d+++++V +V++++u ++++++++V VV+++ +++f+++++VV+VV++++++e++++++VV+++ +++b ++++ ++V++V++++++++g++++++V+VV++++++++c+++++V+V+++ +++d+++++V++++++h++++++%++++++++f+V++++++++++++`+++++++++++++++e+++++++++c++#++++++++++]++"+++++++++++e+++++++++++++++g+ +++++++++++ ++k+++++V+ +++++++++{ +++"+V++++++++o ++++#+V++V++ ++r++++V++VVV+V!V++++++++t ++++++V++VV+!VV++++++++++V5+V+VV+VV+VV+++++w+V"+VV+VV+VV++ ++++FVV++VVVVVV++V+++ +++++++ VV+4VVV+V+V++++++++^+++++++V+VV+V+V++VV++++++ +VV+++ +V&V+VV+V++++VV+V++V +$+++VV++VV++++VVVV++++ V+$+VVVV++++++++V++VV+VVVVVV++++++++V+V VVVV++++++++++ +VVVV VVV ++++++ VV++V+++++VVVVVV+V ++++++ VVVVV+ VVV++++ V,V+VVVVVV++VVV+++++V +OVV++VV+VVVVVV++VV++++V&+VV+++V++5+V++VV+VVV++++++++V{VVVV+V+V++++V++++V+V++V++VV++++++u+VVV++V++V++V+++V+V+VV+++VV++++++4VVV+++V+++VU++V++V+++VV++++V++++++ VQVV+++VV++++++VV+++ +++++ +++cV+V+VV++V++V++V++++++++++++VD++V+V+++V+++++++++++ VV+E+V+V+++VV+V+++V ++++++++++++/V+VVV+VV+V+VV+VV+++V ++++!+++++++]++VV++VVV+VV+V++++ +++++++ +++++9++++VVVVVVVVV+VV+++++++++++++++++++VVVVVVVVVV++++++++++++++++++++++V VVVVVVVV++++++++++++++++++ +++++++++V V VVVVV V+++++++ ++++++++++++++++++++VVVV+VV+++++++++++++++++++++V+V++V+++!++++++++++ +++++++++++++++++V+;++++++++++++++++++++VV+++4V+++VV++++++++++++++++++++++VV+++V++V+V+++++++++++++++++++%+V++++VV+++VV++VV++++++++++++V+VV+V++VVVVV ++++++++ +++++ ++/VV+V++VVVVVVVV+VV++++++++++ + +I++VV+VV+V+++V++V+++++++ ++VVV++F++V+V+++V+VVV++++++++++ +V+++++++V++V+V++V++++++++V+VVF+++V++V+VV+++++++VV++VV5+V+VVV++V+++++++V V+V++VVV++ +++++++V&VVVVV +VV+++++++++++VVVVVV+VV +++++++VV+V++V V++VV+++++VV+V++++V(VV++++++++++++++ +V+++ ++V++++++++++ +++VVVVVVV+VVVVVV++++V+++++++++V+VV++V+V+++++V ##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #d1drw2} ##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse # ddrw2:ddrw2:$d4drw2:@@5 @@( Fundamentald drw2:  *frequency (F0)ddrw2:"#:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #d`drw2m@}o Z ##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse # ##:0 userdict /_cv known {_cv begin stg end}{pop}ifelse #ac{@("ddrw2D++L:BtAl%% S&WordMicrosoft Word&Word   2 @@$Use Word 6.0c or later to 2 @$view Macintosh picture. R1 R1 R1 0I:9"" #K%%DSIDICT:_cv userdict /_cv known not currentdict /bu known and{bu}{save}ifelse userdict /_cv known not dup {userdict /_cv 20 dict put}if _cv begin /bdf{bind def}bind def /isDeviceColor {/currentcolorspace where{pop currentcolorspace 0 get dup /DeviceGray eq exch dup /DeviceRGB eq exch /DeviceCMYK eq or or}{T}ifelse}bdf /setcmykcolor where{/setcmykcolor get /cvcmyk exch def}{/cvcmyk{1 sub 4 1 roll 3{3 index add neg dup 0 lt{pop 0}if 3 1 roll}repeat setrgbcolor pop}bdf }ifelse /stg{isDeviceColor {cf ca /cs load setscreen setgray}{pop}ifelse}bdf /strgb{isDeviceColor {cf ca /cs load setscreen setrgbcolor}{pop pop pop}ifelse}bdf /stcmyk{isDeviceColor {cf ca /cs load setscreen cvcmyk}{pop pop pop pop}ifelse}bdf /min1{dup 0 eq{pop 1}if}bdf currentscreen/cs exch def/ca exch def/cf exch def end currentdict /bn known and{bn}{restore}ifelse  ##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #d0drw20t= } 0t=##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #L\\0t=?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ @`@ @X@ @H@ @@ @@ @@ @@ @8@@pD@@@@@@@@@@@@@ `@@@@@@p@@@@@@@@A@@ A0H8@@ P @@  P<@@@ `@@ @@ `@@ 0@``@@ >@@@@  ``@@0 @@  @@p @@@Ƞ@@ @P@@ @\`@@#`@@ @ @@ x@ @@@@@@@@@@@@@@@@@@@@@@@@@@ @@ @@@ @@@ @ @ @ @ @@ @@ @0@@@@@0@@@@@@@@@@? @@(@ @(9@ @ ?D @@D@@@x D@@@`D @@D@@1 D@@@ D@@@@B@@@B@@@B@@ @ @@@ 9 @@@@@@@0@ @@O@ @@@@ @@ @@  @@ @@ @@ @@px @@ @@ D?pp@@D?@@" D" @@"D0 @@D @@@ D'@@D` @@D  @@H ! @@(" @@($ @@' @ @@@ @@@ @@@ @@ @@@@ @ @ @ @ @ @ @@@0@@0@@P@@H@@H@@@@ @@ `@@ @P@@$<@@@@00@@@ @@` @ @@B @ p @@< @!@ G@!@(A(@%@((D@%@ px(@!@H (@ @ .pH  (@@0  @@@@  @@@1@  @@@@" @@@@@@? @@ @@ ! @@ 9N @@ @@@(B@@@(F@@8@C!@Dp!@d?#@c<&@? ?$@ `'@< '@?? $@8~(@? B)@? L`'C`>'Da$|8 `(X$>0~*`?%@? &@` *@ x2x)@  r+@x? 8'@Ӏ+@00 8~+@  ` +@@p8  x,@  ~,@ 8@`,@8@<` ~-@ 8?pp.@p H`1@  H.@`| 0>`,@ p @`)@p @@*@  `0@)@ |x0 8*@p  `,@ x 2@xp`p1@0 @@,A@` p  0+^ ? @+`p8 0)@  )@ <8 `@@`8)@| 0@0'@ 0'@0 p @.@p`0x0@  /@ @ `  @/@x0 @ p+@ @ ?   +@0 0`  )@`  p@)@ p>`` '@ @ @'@    @ @ -@ @0hF8 80 0/A @H  @@.N @TI@  @0-pG0?x 0,@$A +@$O ,@ |8,@p 0 @)*@@@@xD)@    HF*@ @00I)@p @@@)@ @ @@ 0@F!@@# `1@B 0 @Q@0@ "D@ < Q 0@0.D   /@ ~\p-@x 0 @@,C   @+| ` 00*@   @@*@   0 @*@ `  0)@ @  @)@ @   0*@@   *@@`@`  @)@@@@  @(@0@@ @ )@@ @ @8.@ @@ .@ @@@ 0@0@@ @/@ @@@@ @/@ 8@ @ @1@ @@1@0 0 @` 1@  @ @1@@  @  2@@ @`2@0(@@@2@` @D@@@@2@p0` @@@ 1@ @ D@@@1@ @ D@@@/@@@@@@@@/@0 0@ 0/@ @/@  @@/@<8  @@@/@ @ @@@/C @ @ @0D  ` @a00D @ @@B0D @  @d0H @ @d2X p @2P @? @1` @0` @@1` P 00/@ n @  /@8/@*@ *@@ h*@ @ *@p  @  0*@  @+@  @+@ 0 ` +@@ @+@   @+@8    `+@`@    +@@   +@  +@   *@@ @ @*@0@ @  +@0   +@ @ +A ,F`  ,D@@ *H@ @+P @ @ @@,`p;  ``0h,@ <  ,@ 3 B ,@ 0P b\-@ 3 +@  ` ,@ @`  ,@@@@ ,@``@-@@ ,@0-@.@c(-@  *@<0'@0&@`` "@ "@` "@ @ G0@ 0  X @ ` `   @`0   @@  @@!@@ @0@@@h @@@ @@@0@@@@@ @@  R1##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #d0drw27s=}##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #HH%Gcff33̙ ff 33 ff33ffffffffffff33ff33333333ff333333 !ff"33#$%&'ff(33)*+,̙-ff.33/0̙1̙2̙3̙ff4̙335̙6ff7ff8ff9ffff:ff33;ff<33=33>33?33ff@3333A33BCDEffF33GHIJKffL33MNOP̙QffR33STUVWffX33YZff[ff\ff]ffff^ff33_ff`33a33b33c33ffd3333e33fghiffj33klffmffnffoffffpff33qffrffsfftff̙uffffvff33wffxffyffzff{ffff|ff33}ff~ffffffffffffffffffffff33ffffff33ff33ff33ff33ffff3333ff33ffffffffffff33ff33333333ff333333333333̙33ff33333333333333ff33333333ff33ff33ff33ffff33ff3333ff3333333333333333ff333333333333333333ff333333ff33̙ff33ff33ffffffffffff33ff33333333ff333333ff33ٻڪۈwwUUDD""wwUUDD""wwUUDD""wwwwwwUUUUUUDDDDDD""""""s=v VVVV V  VV VV  VV V    )VVVV VV  V VV VVVV+VV6+VVVVVVVVVVVV VV&VVV VVVV VVVVVV VV VVVVVVVVVVVVV VVVVV   VVV VV VVVVVVVVVVVVVVVVVVVVVVV VVVV VVVV&VVVVVVVVVVVVVVV VVVVVVVVV VVVVVVVVVVVVVV VVVVVVVVVVVVVVV  VVVVVVVVV V VVVVVVVV VVVVVVVVVVV VVV  VVVVVVVVV +VV VVVV VVVVVVV VVVVV VVVVV VVVVVVVVVV VVVVV V  VVVVVVVVVVVVVVV VVVVVVVVV VVVVVVVVVVVV VVV VVVVVVV VVVVVVV+VVVVV VVVVVV VVVV VVV $+VVVVVV VVVVVV VVVVVVVVVVVVVV VVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVV VVVVVVVVVV VVVVVVVVVV 'VVVVVV VVVVVV V VVVVV VVVVVVVV VVVVVVVVVVVVVVVVV  VVVVVV VVVVVVVVVVVVVVVVVVVV VVVVVVVVV V+VVVVVVV VVVV'VVVVVVVVVVV VVVVVV V VVVVVV VVVVVV VVVVVVVVVVVVVVV VVVVV +VVVV VVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV  VVVVVVVVVVVVVV VVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVV VVVV VVVVVVVVVVVVVVVV VVVVV VVVVVVVVVVVVVV V+VVVVVVVVV  VVVV VVVVVV VVVVVV+V VVVVVVVVV +VVVVVVVVVVVVVVVVVV+VVVVV VVVVVV VVVVVVVVVV+VVV VVVVVV+VVVVVVV VVVV VVV VV VVV!VVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVV VVVVVV VVVVVVVVVVV!VVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV VV VVVVVVVVVVVVV VVVVVVVVV VV VVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVV VVV VVVVVV VVVVVVVVVVVVV VVVVVVVVVV VVVVVVVV VVVVVVV VVVVV VVVV VVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV~VVVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVV VVVVV VVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVV!VVVVVVVVVVVVVVVVVVV VVVVVVV V VVVVVVVVVV VVVVVVVVVVVV VV  VVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVV VVVVVVVVVVVV VVVVVVVV VVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVV  VVVVVVVVVVVVVVV VVVVV VVVVVVVVVVV VVVVVVVVVVVV VVVVVVVV VVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVV VVVVVV VV VVVVVVVVVV VVV VVV VVVVV VVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVV VVV VVVVVVVVVV VVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV VVVV VVVVV VVVVVVVVVVVVVV VVVVVVVVVVV VVVVVVVVVVV"V VVVVVVV VVVVVVVVVVVVVVV VVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVV VVVVV VVVVVV VVV VVV VVVVVV VVV VVVVVVVV VVV VVVVVVVVVVV VVVVVVVVVVVVVV VVV VVV VVVVVVVVVVVV VVVVVVVVVV V VVVVVVVVVVVV VVVVVVVVVVV VVVVVV VVVVV!VVVV VVVVVVVVVVVVV + VVVV VVVVV V"V VVVVVVVVVVVVVVVVVVVV VVVVVVVVVVVVVVV&V V VVVVVVVVVV VV VVVVV VVVVVVV% VVVV VVV V VVV VVVV VVV    +  V V VV+VV0 VV +V VVVVVVV VV VV&VVV&VV VVVVVVVVV+VV;V++VVV#VV++V+VVVVV VVVVVVVVVVVV     |V Z##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #d0drw27k1}##:1 userdict /_cv known {_cv begin stg end}{pop}ifelse #HH%Gcff33̙ ff 33 ff33ffffffffffff33ff33333333ff333333 !ff"33#$%&'ff(33)*+,̙-ff.33/0̙1̙2̙3̙ff4̙335̙6ff7ff8ff9ffff:ff33;ff<33=33>33?33ff@3333A33BCDEffF33GHIJKffL33MNOP̙QffR33STUVWffX33YZff[ff\ff]ffff^ff33_ff`33a33b33c33ffd3333e33fghiffj33klffmffnffoffffpff33qffrffsfftff̙uffffvff33wffxffyffzff{ffff|ff33}ff~ffffffffffffffffffffff33ffffff33ff33ff33ff33ffff3333ff33ffffffffffff33ff33333333ff333333333333̙33ff33333333333333ff33333333ff33ff33ff33ffff33ff3333ff3333333333333333ff333333333333333333ff333333ff33̙ff33ff33ffffffffffff33ff33333333ff333333ff33ٻڪۈwwUUDD""wwUUDD""wwUUDD""wwwwwwUUUUUUDDDDDD""""""k1ƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁƁ ƲƵƶƹ V Ƽ VV'ƽ %ƾ8 鬬<VV: ;VCV RV W VV_  YV^鬬VV QVVVRV VхWVVVVX VVVVa VVV X VVVVюiVVVWVVk VVVVy VqV VzVV VVV VVfVcV wV ݁ʁʁVk+V޳ۏݏVuƬVV+VkѠuVVVV+޳kۏr VV+ ߳kwVVVV+ ݏkk kV++      !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrsuvwxyz{|}~ kk܏ށʁ++VkGk݁ V V+++VkV V ޏkGGkkā V V  ݏkk GGk܏VVVVݏkkk ʁV kkľľʁVV ݏkkVVVkkľʁV VV VVV kkkkV ѬVVk kkVkkkk ʁV VV++VkVVV++VVVʁVV+++VVVVV ʁ VVV+VV߳ľV ʁVVV+V V+ ľľľ VVVV܏ľľ V VVVV+VVVVkk܏V+ Vkk܏ľV+V Vݏkk ʁV++Vݏkk V+VV+VVkGGݏľ++V+ V ޏkkG۳܏āV+V+VVVVk܏kkʁ聁VV+++++VVVݏkkkkľ VVVV++VVV޳ݏ޳ݏkĽVV+V++V+++VVVVV+VV+++++VVV V ݏkkkkkVVVV++++VVVVݏ ݏkkľľ+VV+VVV V܏ݏk kk܏޳ľVѼV+VVV ݏkkۏݏkľľ VVV í VV+++++VVkGݏkľľVV++++VVVkkGGݏkkkľʁ V+V+++VkGGkGGGkľVVVV  kG GGkk ʁVVVV+ VVV ߳kGGkkGGkľľVVVVVk kkkGkľVVV+V ݏkGkkGGGk۾ʾVVVVVVVݏ kGGݏkkGGGkľƳ+VVVkݳGGG GGkk܏ľľľVVVkkݏGGGGGkkľľ VV +V kkk kkľľV ++VVV +kkľľ+V+VVV++kkkkޏk kk޾ʬ++VV++ VV ܳkkkʁVV VV+VVVV+ Vkkkkkkkľ ľ VVVVV+VVV++V ľľ VVV+VVVVV VVkľVV+++V+VVV+V܏ľľVVVV+V+++VVV++VVVʁVVV++V+VV+V VVVV++V VVV++V++VľVVVVV+ľVVVVVVV+++ʾľV VVVVV++VVVVľľľVV VV+VľVVV+VV  VоVV++VVV+VVVVVԾVV+V ++V!VVVVVVVVV+VV ++VVVVVVV VV++VVV +VVVV VVVVVVVV VVV + V+VVVV VV+V VVV++ VVV V+