ELF>B@@йAWAVAUATUSH8 L$,8D$D$2H$L$A9t$A9A4,ANXD$|$pH$T$mH=HDH=HH=HH=HH=HH=HH=HH=HH=HH=HH=HH=HH=HH޿ H=HH=HH=HH=HH=HH=HH=HHH1H=HH=HH=HHH1H=HE1D$,D9|$,AGA9DEwXDD$\$D$ f.rf(HDH߰HDH1d$Hf.d$vD$ Xf.vH= E'H޿ H=HHDH1D$HH߰L$Hf.L$vDH=$HHLDl$x1LH=H*H=HL$xHMH1H޿ \$1f.\$v $H8[]A\A]A^A_AWAVAUATIUSEIHH$PHL$@LH|$@t$41LD$8H$HHD$XE$dH%(H$1HHȉD$1A9~-IL$HAE1L CDl$D9AEE1D$D$B09S~bK K2LT$(HHLK L\$(HH P1U|$3EIPDL$ LD$(t$0L\$@H LT$(AIIh|$t!HL$LHIHLH9-tH8H[]A\A]A^A_H8[]A\A]A^A_The NULL Hypothesis: The number of speakers for which the differences is positive equals the number of speakers for which the differences is negative. P(N(+)) = P(N(-)) = 0.50 Alternate Hypothesis: The number of speakers for which the differences is positive is NOT equal to the number of speakers for which the difference is negative. Decision Analysis: Assumptions: A1: The distibution of positive and negative differences follows the binomial distribution for N fair coin tosses. A2: In order to resolve the complication caused by cases where the difference in Word Accuracy is zero, half of the cases will be assigned to N(+) and half to N(-). In the event of an odd number of zero differences, N(-) will get one extra, this reduces the probability of there being a difference between the two systems. Rejection criterion: Reject the null hypothesis at the 95%% confidence level based on the following critical values table. N is the number of speakers being compared and N(-) is the number of negative differences. MIN(N(-),N(+)) P(MIN(N(-),N(+)) | N=%2d) -------------- ------------------------ %3d %5.3f %3d <0.001 <--- Null Hypothesis rejected at or below this point Decision: There were N(-)=%d negative differences , the probability of it occuring is %5.3f, therefore the null hypothesis is REJECTED in favor of the Alternate Hypothesis. Further, %s is the better System. is ACCEPTED There is no statistical difference between %s and %s (Threshold for equal percentages +- %5.3f)%6.2f[%s - %s]Sign Test Calculations Table Comparing%s %s %s Percentages for Systems %s and %s%s Difference Sign0-+Avg. No. Speakers with Positive %s Differences = N(+) = %2dNo. Speakers with Negative %s Differences = N(-) = %2dNo. Speakers with No %s Differences = N(0) = %2d%s.signwWarning: Open of %s for write failed. Using stdout instead. Output written to '%s' SpkrSpeakerComparison Matrix for the Sign Test??Mb@?{Gzt??GCC: (Ubuntu 7.4.0-1ubuntu1~18.04.1) 7.4.0zRx H$ICB D(A0A8Dp8A0A(B BBBlh BBB B(D0A8N] PAZIKKB^s8A0A(B BBBpBBD B(A0A8Gp x\CL\pJ 8D0A(B BBBE D8A0A(B BBB   Z%*/49&>lDJPV\1b}hn tXz"g3W} IV'( .8_c"(.4:@FLRX^d!j pRvZ|\$$ %,8?FO[`qDsigntest.c.LC0.LC1.LC2.LC3.LC4.LC5.LC6.LC7.LC8.LC9.LC10.LC11.LC12.LC13.LC14.LC15.LC16.LC17.LC18.LC19.LC20.LC21.LC22.LC23.LC24.LC25.LC26.LC27.LC28.LC29.LC30.LC31.LC32.LC33.LC34.LC35.LC36.LC37.LC38.LC39.LC41.LC42.LC43.LC44.LC45.LC46.LC47.LC48.LC49.LC50.LC51.LC52.LC53.LC54.LC55.LC56.LC57.LC58.LC59.LC60.LC61.LC62.LC63.LC64.LC65.LC66.LC67.LC68sign_test_analysis_GLOBAL_OFFSET_TABLE_compute_acc_binomialfputsfputc__fprintf_chkcompute_signtest_for_treatment__sprintf_chk_msz_mallocmalloc_diedcenterstrcatrsprintfTEXT_strdupfree__stack_chk_failperform_signteststrcmpstdoutfopenstderr__printf_chkprint_trt_comp_matrix_for_RANK_one_winnerfclose W `Q R RRRRRRR R R'/R6>REMRZSaiRpxRRRRRR T!R"R#R $T%%-R^ lQ Q&'T(T)RS$S+*3R:+LTY,hT-R.T/R0R1TS23VWXWYWX WY 49ZJ[V5[[b4lZy[6V7Z81TB9`\h]xZ8TZ:T4ZR4ZR;R4 ZR;&R-44Z?RF;NRU<]Zd8vT4ZRZR;RZR;RZ R ; R& Z- 8? TJ 4O ZZ Ro S{ S 4 Z R Z R ; R0 T7 ;? Rf Tm ;u R 2 = > Z 8 T ? Z 8 T- @2 Z9 8K T 4 Z R S S 4 Z R A Z R ; R@ TG ;O Rw T~ B R C V Z 8 T D V Z 8 T+ E> VK ZR 8d Tk Bs R} ^ ^ ^ ? @ F O._G\?abHcdITbJeW!X/W4YRWZXiWnYWXWYWXWYxKULMfbg l$D.symtab.strtab.shstrtab.rela.text.data.bss.rodata.str1.1.rodata.cst8.rodata.cst16.comment.note.GNU-stack.rela.eh_frame @@( &S,S12S@@(Mp[0,dyHt@BH  O %`B