AaGހ33 P`PP   @     0ppp0 P`0 HH $d HHHH̀̀̀ff@  'd  Footnote TableFootnote**.\t.\t /:;,.!? .c5 c[ LORUnresolved Cross-RefsLOT TableTitleLOFFigureTOCHeading1Heading2Heading3NAltAug BambergerBiCDsCorCorradaDecDeshmukhDoD Doddington DownArrowFaxFinkeFiscus GanapathirajuGleesonGodfreyGoelGraffHCopyHMMsHTKsHViteHamakerHartwellHemphillHollimanISIPsInstiJun KhudanpurKirchoff LeftArrowManthaMarkovMcDanielMeadeMispronuncations ModellingNockOrdowskiPallett PennsylvaPentiumPicone PrzybockiResegResegmentation RightArrowSaraclarSegmentaSigSimrallSolaris SparcstationsTashaTclTkUpArrowUtt ValidatorWERsWaibelWheatleyWootersWorkflow ZavaliagkosZeng affirmationsb_asidebaseform bookmarking breakpointsbtwncasesensitivecdcepstralcfg check_boundscheck_dictionary check_energy check_lexicon check_silencecheck_speech_rate coarticulatedcoarticulationconfigconfirm_word_files create_configcreate_exciselistcreate_htk_mlfcrossvalidation crosstalkdB delimiterdevdevsetdevtest discluding disfluentdoce_aside echo_canceleducaemailevalevalset excise_signalfaqftp get_val_statsgonnagotta hardlimitedhrshttphttp: implementableinhouseinternetkindaldc linguistcallylogin mentationmetricsmfccmindsetminsmiscapitalizedmisidentificationmistranscribedmlfmrk mrk_to_transmsec multiplepass multispeakernahnalnawnianonmonosyllabic nonspeech non_silence outofbandpaddings perutterance phonebasedpostprocessedpp prepare_dataprev raw_to_nist readme_files realtime recognizer reestimation reframingremapped resegment resegmentatinresegmentations resegmented resegmenting resegments retranscribe retranscribedrogosclite screenshotsecsecs segmenter singlemostsorta spectrogramsplacestateoftheart swb_segmenterthemselftimeconsuming timestamps timeframetimelinetiontranscription_guidelinestriphonetute twochanneltxtuhutterance_hist validators validatorswannawavwaveform waveformswebsite wordbyword workforcewwwxRTyall   Default FontDefault FontEquationVariablesEquationVariablesSymbol/. %  9   N l  "&( F H p\Y  $   #)  i  pppppppppppppppppp# $64828: isip_heading_0: 1. ABSTRACT (65816: isip_heading_0: 2. INTRODUCTION 752168: isip_heading_0: 3. CHANGES IN WORKFLOW PROCESS "19950: isip_heading_0: 5. ISSUES &51715: isip_heading_0: 7. REFERENCES }39328: isip_figure: Figure1. Timeline for the remainder of the SWB resegmentation project. Our anticipated completion date E29256: isip_table_title: Table 1. An overview of the released data. m18033: isip_reference: [1] J.Godfrey, E.Holliman and J.McDaniel, Telephone Speech Corpus for Research and3 544479: isip_heading_0: 5. ANALYSIS OF RELEASED DATA% o33800: isip_reference: [2] B.Wheatley, G.Doddington, C.Hemphill, J.Godfrey, E.C.Holliman, J.McDaniel, and7 -88446: isip_heading_0: 7. ACKNOWLEDGMENTSfs& j96622: isip_reference: [3] J.Hamaker, Y.Zeng, and J.Picone, Rules and Guidelines for Transcription and. m96443: isip_reference: [2] J.Godfrey, E.Holliman and J.McDaniel, Telephone Speech Corpus for Research and' j13250: isip_reference: [3] J.Hamaker, Y.Zeng, and J.Picone, Rules and Guidelines for Transcription and h15240: isip_reference: [5] J.Hamaker and J.Picone, The SWITCHBOARD Frequently Asked Questions(FAQ), l94546: isip_reference: [6] J.Hamaker, A. Ganapathiraju, and J.Picone, SWITCHBOARD Educational Resources, s77400: isip_reference: [7] J.Hamaker and J.Picone, A Statistical Guide to SWITCHBOARD: Topic Statistics, http:/    7NP X:m96443: isip_reference: [2] J.Godfrey, E.Holliman and J.McDaniel, Telephone Speech Corpus for Research and7NP Xo33800: isip_reference: [2] B.Wheatley, G.Doddington, C.Hemphill, J.Godfrey, E.C.Holliman, J.McDaniel, and7NP Yj96622: isip_reference: [3] J.Hamaker, Y.Zeng, and J.Picone, Rules and Guidelines for Transcription and7NP [h15240: isip_reference: [5] J.Hamaker and J.Picone, The SWITCHBOARD Frequently Asked Questions(FAQ),7NP \l94546: isip_reference: [6] J.Hamaker, A. Ganapathiraju, and J.Picone, SWITCHBOARD Educational Resources,7NP ]s77400: isip_reference: [7] J.Hamaker and J.Picone, A Statistical Guide to SWITCHBOARD: Topic Statistics, http:/7NP ^E29256: isip_table_title: Table 1. An overview of the released data.7NP _m18033: isip_reference: [1] J.Godfrey, E.Holliman and J.McDaniel, Telephone Speech Corpus for Research and7NP ` }39328: isip_figure: Figure1. Timeline for the remainder of the SWB resegmentation project. Our anticipated completion date7NP b j13250: isip_reference: [3] J.Hamaker, Y.Zeng, and J.Picone, Rules and Guidelines for Transcription and <$paratext[Title]> <$paratext[Heading1]> <$curpagenum><$lastpagenum>"<$monthnum>/<$daynum>/<$shortyear><$monthname> <$daynum>, <$year>;<$monthname> <$daynum>, <$year> <$hour>:<$minute00> <$ampm>"<$monthnum>/<$daynum>/<$shortyear> (Continued)+ (Sheet <$tblsheetnum> of <$tblsheetcount>) <$marker1> <$marker2> isip_name/INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING<$monthname> <$daynum>, <$year>"<$monthnum>/<$daynum>/<$shortyear> <$fullfilename> isip_paratext <$paratext>isip_toc<$paratext>\t\t<$pagenum> isip_figureFigure <$paranumonly>isip_ref[<$paranumonly>] isip_paranum<$paranumonly> <$filename>isip_document_name2Improved Monosyllabic Word Modeling on SWITCHBOARDisip_document_dateNovember 15, 1998ooqqAttvvCxxzz isip_title:: isip_preface  TOCIXLOFLOT#HTML&&Headings6699 isip_preface>>AUC* X'k@&kB& P& Z' ^' V' `' ' \' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' '*' '*' '*'k&k&k& & ' ' ' ' ' ' ' 'k&k& & '' '''''''''''''''''''''''*''*' E'*'l&l&l& ;& G' I' M' C'&' Q' ' K' ' ' ' '&' ' ' ' ' ' ' '' ' ' ' ' ' ' ' ''' ' ' ' ' ' ' 'l&l& & ' ' ' ' W' ' Y' 3' U'l&l&l&l& '& 5' 7' 9' =' 1' ?' A' C' G' ;' I' K' M'"' E' ' ' '"' ''j''l' ' ' ' ' ''~''' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' '#' ' }' ' '#' y'(R'(T' s' u' w'#' o'(f'(h' i' k' m' ]' e'(z'(|' _' a' c' S' [' U' W' Y' I' Q' K' M' O' ?' G' A' C' E' 5' =' 7' 9' ;' +' 3' -' /' 1' !' )' #' %' '' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' '*' '*'*'*' v'*'o&o& n& x'%( t'''m)#'o'g*k'i'a+e'c'[+_']',Y'',' ' '' ' ' ' ' ' ' ' ' ' '9- '9@6>6=6172.E2556A619B9E9K9L. i- K-888871.8I39?61{783.88%P[5]P<[7]P<[6]P<[8]MMMMM M M P<[1]MMMMMMN Table 1.P<[2] Figure1..448777.P<[4]8888P[3]a88!444444>4>5>43434>4>4>4>4>444+D4+E6+F4+G4+H4+I4+J4+K44448888887!785.B87'74.76. 9C9D9F9G9H9I9J;8=/?9@:A0B8;d9otd6vzqG p>45" upw>upwK =FrameImage #%vMLYj}۶mI$۶mI$۶mI$. K   L     ytxytxy       ,   X_ i[`j^cktck                 ae@+aenaenae9           $   )mbfmbfqbfnb"    '   %     : .(e_ cdlcdlc`     "?      <      > !  ? !  ?!  ? ! .?N! .?N!b.?L"a.3I#` (.   >G   >H ,H  +H  + H  , ,H.>_ ,H.?_ +H(@/-)I#/-  I#. O4# / $O5#A $O5#A $B6#A $B3" $C2! $F V#G VR< S:/ 8- 7  + 6  ) 4  &   2  '&  1 & %  1 %# 0 %$ / +# 0 "# 0  " %  0    "  / !  /  !  .      .    !  -     -     0     ,                   I    K  K  M G > !> #> (>#,>)+? -+?5-J8- @?/Y!G  0[l1\n1]p2^rS_tU`wWb9 X6 )  P@ 5  =EndInset }$ k=!.$ "WeCharacter Macros }l k?!l "We } kA! "We }$HkD! $H"W e Character }lkF! l"W e Replace With }kH! "W e Comments }$HkM! $H"W  e }lkO! l"W e¢ }kQ! "W e }$HkV! $H"W  e }lkX!l"W  e© }kZ!"W  e }$Hk_!$H"W e }lka!l"W  e® }kc!"W e }$Hkh!$H"W e }lkj!l"W e° }kl!"W e }$.Hkq!$.H"W e }l.ks!l."W e-- }.ku!."W e }$>Hkz!$>H"W e }l>k|!l>"W e- }>k~!>"W e }$NHk!$NH" W e }lNk!lN" W e... }Nk! N" W e }$^Hk!!$^H" W e }l^k! "l^" W e }^k!!^" W e }$ k!O'$ " W#eGeneral Macros } k! " W e } k! " W!e }[ k![ " W"e }$gk!#($g" W' e Macro Name }gk!')g" W e Replace With }gk!(*g" W$ eHead }[k!)+[" W% e Comments }$gk!*,$g" W+ e }gk!+-g" W& e }gk!,.g" W( e }[k!-[" W) e }$` kw2$` W.eCross-Reference Macros }` k` W,e }D` kD` W-e }$pk/3$pW1 e Macro Name }pk24pW* e Replace With }Dpk35DpW/ e Comments }$€k46$€W4 e Figure All }€k57€0 %Figure <$paranumonly>, PE <$paratext> }D€k68D€W2 e }$žk79$žW7 eFigure Number }žk8:žW3 eFigure <$paranumonly> }Džk9;DžW5 e }$®k:<$®W: eHeading }®k;=®W6 e <$paratext> }D®k<>D®W8 e }$¾k=?$¾W= e See Also }¾k>@¾W9 eSee <$paratext>. }D¾k?AD¾W; e }$k@B$W@ e Section All }kAC< %Section <$paranumonly>, PE <$paratext> }DkBDW> e }$Dk!E$D"WC eSection Number }Dk!DFD"W? eSection <$paranumonly> }DDk!EGDD"WA e }$Tk!FH$T"WF e Table All }Tk!GIT"B %Table <$paranumonly>, PE <$paratext> }DTl!HJDT"WD e }$rl!IK$r"WI e Table Number }rl!JLr"WE eTable <$paranumonly> }Drl !KMDr"WG e }$l!LN$"WL e }l!MO"WH e }Dl!N#D"WJ e }$J lT$J WPeSystem Macros }J lJ WMe }J lJ WNe }[J l[J WOe }$Zgl PU$ZgWT e Macro Name }Zgl"TVZgWK e Replace With }Zgl$UWZgWQ eHead }[Zgl&VX[ZgWR e Comments }$jg,l,WY$jg,WX e StartOfDoc }jg,l.XZjg,WS e }jg,l0Y[jg,,U % ��<$defaulttitle></P��ETITLE> }[j��g,���l2�Z\��[j��g,�WV ��e }$��g���l8�[]��$��g��W\ ��e EndOfDoc }��g���l:�\^����g�WW ��e }��g���l<�]_����g�WY ��e }[��g���l>�^`��[��g�WZ ��e }$��g,���lD�_a��$��g,��W` ��eStartOfSubDoc }��g,���lF�`b����g,�W[ ��e }��g,���lH�ac����g,,�] ��%<TITLE> ��<$defaulttitle></P��ETITLE> }[��g,���lJ�bd��[��g,�W^ ��e }$��g���lP�ce��$��g��Wd ��e EndOfSubDoc }��g���lR�df����g�W_ ��e }��g���lT�eg����g�Wa ��e }[��g���lV�fh��[��g�Wb ��e }$��g,���l\�gi��$��g,��Wh ��eStartOfFirstSubDoc }��g,���l^�hj����g,�Wc ��e }��g,���l`�ik����g,,�e ��%<TITLE> ��<$defaulttitle></P��ETITLE> }[��g,���lb�jl��[��g,�Wf ��e }$��g���lh�km��$��g� �Wl ��eEndOfFirstSubDoc }��g���lj�ln����g� Wg ��e }��g���ll�mo����g� Wi ��e }[��g���ln�np��[��g� Wj ��e }$��g,���lt�oq��$��g,�!�Wp ��eStartOfLastSubDoc }��g,���lv�pr����g,�!Wk ��e }��g,���lx�qs����g,,�!m ��%<TITLE> ��<$defaulttitle></P��ETITLE> }[��g,���lz�rt��[��g,�!Wn ��e }$J��g���l�su��$J��g�"�Wt ��eEndOfLastSubDoc }J��g���l�tv��J��g�"Wo ��e }J��g���l�uw��J��g�"Wq ��e }[J��g���l�v/��[J��g�"Wr ��e }$���� ���l�F{��$���� �#�Ww��eHTML Options Table }������ ���l����������� �#Wu��e }h������ ���l�����h������ �#Wv��e }$�������l�x|��$�����$�Wz ��eControl }���������l�{}���������$Ws ��eValue }h����H���l�|~��h����H�$Wx ��e Comments }$�������l�}��$�����%�W} ��e Image Format }���������l�~����������%y ��% 0001IMAGGIF P��E MACP0001GIF }h����H���l���h����H�%W{ ��e }$4�������l����$4�����&�W� ��e!Copy Files Imported by Reference }��4�������l�����4�����&W| ��eN }h��4��H���l�P��h��4��H�&W~ ��e }$6�� ���l�� ��$6�� �'�W��eHTML Mapping Table }��6�� ���l�������6�� �'W��e }��6�� ���l�������6�� �'W��e }2��6�� ���l�����2��6�� �'W��e }h��6�� ���l�����h��6�� �'W��e }$F��.���l� ��$F��.�(�W  ��eFrameMaker Source Item }��F~���l� ����F~�(W ��e HTML Item }��F6���l�������F6�(W ��e }2��F6.���l� ��2��F6.�( ��%Include P��EAuto# }h��F��.���l� ��h��F��.�(W ��e Comments }$V�����l�����$V���)�W ��e }��VH���l� ����VH�)W  ��eElement }��V6���l�����V6�)  ��%New Web P��EPage? }2��V6���l�����2��V6�)W  ��e }h��V�����l�����h��V���)W  ��e }$t�����l���$t���*�W ��eP:Body }��tH���l�����tH�*W ��eP }��t6���l�����t6�*W ��eN }2��t6���l���2��t6�*W ��eN }h��t�����l���h��t���*W ��e }$����,���l���$����,�+�W ��e P:Bulleted }����H,���l�������H,,�+ ��eLI ��e Parent = UL Q��e Depth = 0 }����6,���l�������6,�+W ��eN }2����6,���l���2����6,�+W ��eN }h������,���l���h������,�+W ��e }$����,���l���$����,�,�W  ��eP:BulletedCont }����H,���l�������H,,�, ��eP !��e Parent = UL Q"��e Depth = 0 }����6,���l� ������6,�,W ��eN }2����6,���l�!��2����6,�,W ��eN }h������,���l� "��h������,�,W ��e }$�������m�!#��$�����-�W' ��e P:CellBody }����H���m �"$������H�-W ��eP }����6���m �#%������6�-W# ��eN }2����6���m�$&��2����6�-W$ ��eN }h���������m�%'��h�������-W% ��e }$�������m�&(��$�����.�W, ��eP:CellHeading }����H���m�')������H�.W& ��eP }����6���m�(*������6�.W( ��eN }2����6���m�)+��2����6�.W) ��eN }h���������m�*,��h�������.W* ��e }$�������m&�+-��$�����/�W1 ��e P:Equation }����H���m(�,.������H�/W+ ��eP }����6���m*�-/������6�/W- ��eN }2����6���m,�.0��2����6�/W. ��eN }h���������m.�/1��h�������/W/ ��e }$ �������m5�02��$ �����0�W6 ��e P:Extract }�� ��H���m7�13���� ��H�00 ��%BLOCKQUOP��ETE }�� ��6���m9�24���� ��6�0W2 ��eN }2�� ��6���m;�35��2�� ��6�0W3 ��eN }h�� �������m=�46��h�� �����0W4 ��e }$*�������mD�57��$*�����1�W; ��e P:Figure }��*��H���mF�68����*��H�1W5 ��eH* }��*��6���mH�79����*��6�1W7 ��eN }2��*��6���mJ�8:��2��*��6�1W8 ��eY }h��*�������mL�9;��h��*�����1W9 ��e }$:�������mS�:<��$:�����2�W@ ��e P:Footnote }��:��H���mU�;=����:��H�2W: ��eP }��:��6���mW�<>����:��6�2W< ��eN }2��:��6���mY�=?��2��:��6�2W= ��eN }h��:�������m[�>@��h��:�����2W> ��e }$J�������mb�?A��$J�����3�WE ��e P:Heading1 }��J��H���md�@B����J��H�3W? ��eH* }��J��6���mf�AC����J��6�3WA ��eN }2��J��6���mh�BD��2��J��6�3WB ��eN }h��J�������mj�CE��h��J�����3WC ��e }$Z�������mq�DF��$Z�����4�WJ ��e P:Heading2 }��Z��H���ms�EG����Z��H�4WD ��eH* }��Z��6���mu�FH����Z��6�4WF ��eN }2��Z��6���mw�GI��2��Z��6�4WG ��eN }h��Z�������my�HJ��h��Z�����4WH ��e }$j�������m�IK��$j�����5�WO ��e P:Heading3 }��j��H���m�JL����j��H�5WI ��eH* }��j��6���m�KM����j��6�5WK ��eN }2��j��6���m�LN��2��j��6�5WL ��eN }h��j�������m�MO��h��j�����5WM ��e }$z�������m�NP��$z�����6�WT ��eP:HeadingRunIn }��z��H���m�OQ����z��H�6WN ��eH* }��z��6���m�PR����z��6�6WP ��eN }2��z��6���m�QS��2��z��6�6WQ ��eN }h��z�������m�RT��h��z�����6WR ��e }$�������m�SU��$�����7�WY ��eP:Mapping Table Cell }����H���m�TV������H�7WS ��eP }����6���m�UW������6�7WU ��eN }2����6���m�VX��2����6�7WV ��eN }h���������m�WY��h�������7WW ��e }$�������m�XZ��$�����8�W^ ��eP:Mapping Table Title }����H���m�Y[������H�8WX ��eP }����6���m�Z\������6�8WZ ��eN }2����6���m�[]��2����6�8W[ ��eN }h���������m�\^��h�������8W\ ��e }$����,���m�]_��$����,�9�Wc ��e P:Numbered }����H,���m�^`������H,,�9] ��eLI d��e Parent = OL Qe��e Depth = 0 }����6,���m�_a������6,�9W_ ��eN }2����6,���m�`b��2����6,�9W` ��eN }h������,���m�ac��h������,�9Wa ��e }$����,���m�bd��$����,�:�Wj ��e P:Numbered1 }����H,���m�ce������H,,�:b ��eLI k��e Parent = OL Ql��e Depth = 0 }����6,���m�df������6,�:Wf ��eN }2����6,���m�eg��2����6,�:Wg ��eN }h������,���m�fh��h������,�:Wh ��e }$����,���m�gi��$����,�;�Wq ��eP:NumberedCont }����H,���m�hj������H,,�;i ��eP r��e Parent = OL Qs��e Depth = 0 }����6,���m�ik������6,�;Wm ��eN }2����6,���m�jl��2����6,�;Wn ��eN }h������,���m�km��h������,�;Wo ��e }$.�������m�ln��$.�����<�Wx ��eP:ReportAuthor }��.��H���m�mo����.��H�<Wp ��eP }��.��6���m�np����.��6�<Wt ��eN }2��.��6���m�oq��2��.��6�<Wu ��eN }h��.�������m�pr��h��.�����<Wv ��e }$>�������m�qs��$>�����=�W} ��eP:ReportTitle }��>��H���n��rt����>��H�=Ww ��eH* }��>��6���n�su����>��6�=Wy ��eN }2��>��6���n�tv��2��>��6�=Wz ��eN }h��>�������n�uw��h��>�����=W{ ��e }$N�������n �vx��$N�����>�W ��eP:TableFootnote }��N��H���n�wy����N��H�>W| ��eP }��N��6���n�xz����N��6�>W~ ��eN }2��N��6���n�y{��2��N��6�>W ��eN }h��N�������n�z|��h��N�����>W� ��e }$^�������n�{}��$^�����?�W ��e P:TableTitle }��^��H���n�|~����^��H�?W ��eH* }��^��6���n �}����^��6�?W ��eN }2��^��6���n"�~���2��^��6�?W ��eY }h��^�������n$���h��^�����?W ��e }$n�������n+����$n�����@�W  ��eC:BulletSymbol }��n��H���n-�����n��H�@ ��%THROW P��EAWAY }��n��6���n/�����n��6�@W ��eN }2��n��6���n1���2��n��6�@W  ��eN }h��n�������n3���h��n�����@W  ��e }$Œ�������n:���$Œ�����A�W ��e C:Callout }��Œ��H���n<�����Œ��H�AW  ��eEM }��Œ��6���n>� ����Œ��6�AW  ��eN }2��Œ��6���n@� ��2��Œ��6�AW ��eN }h��Œ�������nB� ��h��Œ�����AW ��e }$œ�������nI� ��$œ�����B�W ��e C:Emphasis }��œ��H���nK� ����œ��H�BW ��eEM }��œ��6���nM� ����œ��6�BW ��eN }2��œ��6���nO� ��2��œ��6�BW ��eN }h��œ�������nQ���h��œ�����BW ��e }$¬�������nX���$¬�����C�W ��eC:EquationVariables }��¬��H���nZ�����¬��H�CW ��eEM }��¬��6���n\�����¬��6�CW ��eN }2��¬��6���n^���2��¬��6�CW ��eN }h��¬�������n`���h��¬�����CW ��e }$¼�������ng���$¼�����D�W  ��e X:Figure All }��¼��H���ni�����¼��H�DW ��e Figure All }��¼��6���nk�����¼��6�DW ��eN }2��¼��6���nm���2��¼��6�DW ��eN }h��¼�������no���h��¼�����DW ��e }$�������nv���$�����E�W% ��eX:Figure Number & Page }����H���nx�������H�E ��%Figure P��ENumber }����6���nz�������6�EW! ��eN }2����6���n|���2����6�EW" ��eN }h���������n~����h�������EW# ��e }$b�����n�� ��$b���F�W* ��eX:Heading & Page }��bH���n�!����bH�FW$ ��eHeading }��b6���n� "����b6�FW& ��eN }2��b6���n�!#��2��b6�FW' ��eN }h��b�����n�"$��h��b���FW( ��e }$r�����n�#%��$r���G�W/ ��eX:Page }��rH���n�$&����rH�GW) ��eHeading }��r6���n�%'����r6�GW+ ��eN }2��r6���n�&(��2��r6�GW, ��eN }h��r�����n�')��h��r���GW- ��e }$�������n�(*��$�����H�W4 ��eX:Section All }����H���n�)+������H�HW. ��e See Also }����6���n�*,������6�HW0 ��eN }2����6���n�+-��2����6�HW1 ��eN }h���������n�,.��h�������HW2 ��e }$�������n�-/��$�����I�W9 ��eX:Section Number & Page }����H���n�.0������H�IW3 ��e See Also }����6���n�/1������6�IW5 ��eN }2����6���n�02��2����6�IW6 ��eN }h���������n�13��h�������IW7 ��e }$�������n�24��$�����J�W> ��eX:See Heading & Page }����H���n�35������H�JW8 ��e See Also }����6���n�46������6�JW: ��eN }2����6���n�57��2����6�JW; ��eN }h���������n�68��h�������JW< ��e }$�������n�79��$�����K�WC ��e X:Table All }����H���n�8:������H�KW= ��e Table All }����6���n�9;������6�KW? ��eN }2����6���n�:<��2����6�KW@ ��eN }h���������n�;=��h�������KWA ��e }$�������n�<>��$�����L�WH ��eX:Table Number & Page }����H���n�=?������H�LB ��%Table P��ENumber }����6���n�>@������6�LWD ��eN }2����6���n�?A��2����6�LWE ��eN }h���������n�@B��h�������LWF ��e }$�������n�AC��$�����M�WM ��e }����H���n�BD������H�MWG ��e }����6���n�CE������6�MWI ��e }2����6���n�DF��2����6�MWJ ��e }h���������n�Ex��h�������MWK ��e }$6�� ���n�$�J��$6�� %�N�WP��eHeadings Table }��6�� ���n�$������6�� %�NWN��e } ��6�� ���o�$���� ��6�� %�NWO��e }$Fl���o�$GK��$Fl%�O�WS ��eHeading Level }��F�����o�$JL����F��%�OWL ��eParagraph Format } ��F�����o�$KM�� ��F��%�OWQ ��e Comments }$Vl���o �$LN��$Vl%�P�WV ��e1 }��V�����o�$MO����V��%�PWR��e ReportTitle } ��V�����o�$NP�� ��V��%�PWT ��e }$rl���o�$OQ��$rl%�Q�WY ��e2 }��r�����o�$PR����r��%�QWU��e Heading1 } ��r�����o�$QS�� ��r��%�QWW ��e }$l���o�$RT��$l%�R�W\ ��e3 }�������o!�$SU������%�RWXUTUT��e Heading2 } �������o#�$TV�� ����%�RWZ ��e }$l���o(�$UW��$l%�S�W_ ��e4 }�������o*�$VX������%�SW[ ��e Heading3 } �������o,�$WY�� ����%�SW] ��e }$l���o1�$XZ��$l%�T�Wb ��e5 }�������o3�$Y[������%�TW^ ��e HeadingRunIn } �������o5�$Z\�� ����%�TW` ��e }$l���o:�$[]��$l%�U�We ��e6 }�������o<�$\^������%�UWa��eFigure } �������o>�$]_�� ����%�UWc ��e }$l���oC�$^`��$l%�V�Wh ��e6 }�������oE�$_a������%�VWd��e TableTitle } �������oG�$`b�� ����%�VWf ��e }$l���oL�$ac��$l%�W�Wk ��e }�������oN�$bd������%�WWg ��e } �������oP�$ce�� ����%�WWi ��e }$l���oU�$df��$l%�X�Wn ��e }�������oW�$eg������%�XWj ��e } �������oY�$fh�� ����%�XWl ��e }$l���o^�$gi��$l%�Y�Wq ��e }�������o`�$hj������%�YWm ��e } �������ob�$ik�� ����%�YWo ��e }$l���og�$jl��$l%�Z�Wt ��e }�������oi�$km������%�ZWp ��e } �������ok�$l��� ����%�ZWr ��e H I��K���op�����H I��K��������o����W UTUT��m<�F#�G�HApril 2, 1999�I�JRunning H/F 1�K H I��K�� <��q��H I��K��nn ����l��� HH��‘�����os�����HH��‘�� ��������q����Ws��e HH��‘���� =�or��HH��‘����pp ����l���H������� �qt��H������H������ H$��1&���ow�����H$��1&��������t����WuUTUT��e H$��1&�� >�r���H$��1&��ss ����l��� HH��ˆ�����oz�����HH��ˆ����������v����W UTUT��e HH��ˆ����6��x��HH��ˆ����uu ����l��� HUa������o}�����HUa�����������x����WwUTUT��hL�>INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING�?FEBRUARy 15, 1999 HUa�����6�vz��HUa�����ww ����l��� H$������o�����H$�����������z����WxUTUT��lg�@Improved Monosyllabic Word Modeling on SWITCHBOARD�APage �B4�C of�D4�E H$�����6�x���H$�����yy ����l�����d������6��������}:��������SuLqZ�����+E���SuLqZ���SuLquLqC�C���‘����<�{�~��C�C���‘��HH��ˆ����=�{}��HH��ˆ��A´U����>�{~���A´U��c6$��?�{��c6$� ®D]HH��@�{���®D]HH��z���� =FrameImage #%v��Yj��������������������!)9BJcks{kJ֥skRZJ9ZRB91!!1kZR1cB!9sRJ޵sZkB)BcB{9J11!1!kJ)Z{sJZB91!)skRJc{R1JZRcZRJB19!1)!{ZB9kcBk!BJ1)!{skZRJZcB9)!s{kRkZcRZR1BJ))޽{Z9R1)έkcBkcJ!ZB!RֵBsJ91!kZֽkJ����!)9BJcks{{sޭ{sZcRBkcRJB1)9c޽skJ){Z9c֜{s9ƥkΥƽc)J{!sRZ{Rs9Rc1J{Z9{{BZskc{Jsc{Js)kcJsZRc9s1kcRΜ{sk{9kZB1{Z{RZƥcBcBB!ֵskJ{sZ){Z1{19)޽JZJB1scֽkJ����!)9BJcks{skޭ{sZcRBkcRJB1)9k{sR1cBs筌Jέs޵s1Z){ZkcJkBccBΥΜZ{{sΔZƽc9kƥkc{JBs絭ƵRcJƜֵέƽ{sέkJsRZ1޽{sƭR֌k1sBJZBƥZƥ{scƥsR����{����€����v������������w���v�������€�v����T�� ��T����������� �������¦��������ƙw��������w�������̀��������������� �������� ������������逡��€�T������ ���������ހ����������ƀ ���������������€������������ ������������������� �������������w����€ �����w���������逌���T���� �+�������������������7�����������逋�������7������������̀�����������9�������������w����€<�������w���̀�������������R���{���������wX����F �l��w��v������\���� �k������f��� �j������h��l�v��̀k� ��k�ƀ����w�d��������x��ƀ^����}�T������^�������}����wT]�������~����Z��2�FTTTGY֑wUY4����Z��+�L��L��s�axwBUT0�������[��w�%�FTjsssss CaaaaaaxCwGT,�������U�����"�Lssss�sss�aaaeCwT,����T���� � FY]�]~~��ssssss�s�aaaaaaG(����̀��H���ujssssss]]~~ssss�sssaaxzzzz?zw&��������F��������Yssss�s~~]sss�ss�s�aaaaaeeezC3̀(���F����FK� ss]sss~~s~sss��aaaa?aaaa^z3GT$���TF��~~]~~]~~�ss�]��s��ssaa슏arx"������G��FT~~~~~~~]�ssss]~~]�L!vw]�����������aaaaarrrxT!�����C���T����uj��]~~~~~Lj�~�� aaB왊arrrr?xx#����C��̀�YY�j~R~c�j~jj~ss]~sjs~�s~��saaaa왨rrrOx !����C�����|jLj~jj~~s~ss]�ss���aaaaeaazzrrrr^\3 ��C�T����ӥ]~]Lc~�s~sss]ss���aaaaaae^rrrr{xx�C�����Y~]j~R�sjsssss��aaaezފr{{x� �/����Fڜjjj~]~s��sssss��ssssaaaaaa\rOr7C� �-����;o]�j]~sjssssaCaaaarr^zrrOr{z{x� �,������ӑjYooj�_sssss��sssaaaaaaerr{^\⭊^z'�.�w�����K_jjj��oj|]ccjssssss~ssaaxaaaaar^Qr^\%�0�����Ejj__oo_jjY]jss�ssaaaaarrrrQ{rr\\O"�2�����ڜo�_��|jssss�ssaaaaar⊀rOOr\r{z �4�����|_j_m_oo__|Rjoj~~~]Y~~~]~~]~]]~�~ݧ�aaaeaerrrr^ar{\r\O{�6�����;j__j~oom_ojjA|_j]sssss�s��������aaaaaerrr{{\Q��1�w�FYoj___o|omoo_|jssssssL?arrrrOr⭭rޭr{΀��2��|j_ooR|m__o_ojccsssss�ssazCzrrrrOrrr⭭`8Q����4��|_Ro___m_||_o|jcsssssCaaaaaaBCOrrrrrrr7^\Q ��ƀ3��Kjo__o_mRmm|jcssssss�aaaaezފrr?rrr^`8z��3��FL__oځoomojssss�s�aaaaaaaaza^rrrr^`Q\΀��C�ף__m_v|moooo|o_ojcssssaae?rr?Orrrrr\QQƀ�€�B�R__||oooommooomjcsss�s�aaaaaaa\rrrrOOr^˭r7`��A�uY__omoommoomoommoo�cssssss��aae^^rrOOO``QɭU�T����8�u___੣oooooomojssss�aaaaaٿrrOO\`Dfx����6�_o|oooommoojcccs�aaaaaaaaaް^?rrOr\rD`fx��4�EK]__o|oomomjccsssssaeaްarrrrrrr\V```VH��T�3�KRo_omoopomo_]ss�aaaaaaaerr^rrrr\7˭``˔H��6�Ro__|oo oqjcsssssBaaaa?e^?rrrr^DQH�T5�_Rjo_YmcccssssxsaaaaaerrOrrr`r`rD`ʔx���6�u�o_oYooocsssssss��aaaaerrOrrrr`Q`Q<U�T:�om__o|omocccsss��a?aaaerrrrrOr`d����w1�F|�jomqqqcs�saaaaaerr?^rr`D`DX{ƀ�T���/�JR_ooqcsssssss���aaaaerrr^arr`r\DD`Q˔9΀�L.�ͩo_j__ompqccssssss���aaaaarr^?``D=I�w���,�Lo__|ooqqqocsss�]w�aaaaaaerrrrr?````X{77��T,�o |o}qqccs]~]�ss�s�aaaaaaerrrrrr\r```D7H��+�__oommqqqqq|cc~~~]�s�ssaarrrrrrr\````dᱛXd��*�F__|oqqqqq|o~o~s�saaaaerr⭭r\DD``DH΀�vT*�6Ԁ�___ooo} ccssss�ssss�aaerrr\``D`DᱞHd̀�T+�_|ool}lpqqqqcss~sssaaaaaeerrrrrrr\`DD`Hd��%�]_omll}}}lllqqqqqqqqcccss�saaaerrrr\`O`DDdឱI=G��w$�絑___ooolyll}[[}}}lqq cs]sss�aaaaerrrr\D```D`ddě�"�__oollly}}}[NZN[[}}}lllpqqqq|csssssaaaerrrrrrr^`DDDDddٌ7�ẁ#�jooool}[yZ[llqqqqccsssaaaaeerr^`D``Ddd==� �Rԟ__oolll[y[Zhh׀Z}}llpqqqqqqcccs�ssa�aaaerrrrrr\r`D`ឱdV9€���T�Y]_oo_oomll}[yZZZhhhJhhhZZZ[[}llpqqqqqq|csss]ssaaaerrrrD`D`Dddٌ���ן__ooll}[yyMhhh׀JMhZZ}llpqqqqccsss�saaaerrr^^`D`ᱞddfÀ�(�~j_ooll[yhyMhJ@JWW@@JJhhhZZ}llqqqccs]sssa�aaeerrrrr``D`d`ddd=I���v_j_oll}[hyhJJJW44;44WWJJhhZZ[llupqqqqcss�saaaerrrr`r<D``ᱞI=Ѐ�w�矑__oool}[yZhJ@W44;4WJJhhZZ[llqqqccsss]ssaaerrrrr⭭r`Q`D`Ddd=۞��~~~Roool}}yh5hJJW4;6;4WWJhhZk}llbqqcss~s�arrrr\```ᱞd=ی<��F|�|_ol}5ZhhJW4;66666;4WJhhZZllpbqqqqq|css�ssaaerrrrrrDޔ`D`dd=۞΀��|~~~_jooll}yyhhJJ4;66666;4WJhhZZ}lluppbqqqqqss~ssssaaerrr ۭ``d=dŌ�����׈~]o_ol[yMhJW466E6664WWJhZZlubqqqq|s]ssaerrrrrrrD``Q`dd==!��w�jL~j~oollyyZhJJ446E66;4JJhZZllbpqccss~sssarr```ddd=�v�׈j~j_ool[yyZMJJ4;6���6;4@JhMZkl}pqqqqs~sssssaaa쏏eerrrrrDD``Dឱd=d>À��j�~_olykyMhJJ4;6���6664WhhZZ}l}bq|sss~sssssaaaaeerrrrr \`QDdd=d۞��Fs~�]ooo[}[ZZh@4;6��E6;4WJhhZZkkllqqqs~ssssaaaaerrrr``Dᱞ`d㞌=!�T�j~joy}[ZZJW4;6666;WJJhZZklll |css]ssa?arrrr`Ddddd=dd=I�T���̀�Fs~~o[ll}yZMJJW46664W@JMZNkl}ulu}qqqqqqqYjj~s~ssaaaaerrrrrD`D``Qd=㞌=΀����~oo}[ZhhJW4;6EEE666;4WJhhyyy[}pqqqq__~j~j~j~jssssaaa?arrrrrD``Dd=d㞞ـ��͈~~~joo}ll}[ZZhJW4;666;WFJMZkkbpppqqqcssL~�saaaaerrrrrrr``D```dddd㞌ی>�wT�~�]ݠ_ool}}[ZhhJJW4;666 6;4WJ@hMZNllnbbqq|�s�Baaa?rrrrrrD`ᖛ=ߌЀ�̀�~~�oooool}[yZhhJF4;6;;;44WWJhhMZkknbqqqqqcss~ssss?aerrrrD`D``dʞd㞌I��Fsjs~]ool}Z5JJ@WW4WW@hJMMZkkllpqq|s~ssaCCCrr```D`Qʔddd㞞f!��~]sjjoo}yZZyhhJJJWWWJ@JJJhMMZNklbbbqqqq|s~sssaa?xCOrrr`DD<dd㌌9V���;s~�~o}l}}kyZZhMhJMhhMNZklnppqqqqcs]ssssaaaaaaC^rr ```DdX^dd㞌f���� �s~~ooooul}k[ZZZhMhJJMhMMMZZNkkknlbpbqqqqqoocs~ssaaar?٨rrrrrD```=ddd㞌fπ�� �j~~��oool}k[ZMMMZMZZZNkklppqqqqqs~saaaaaarr^zٔr````ddۀ��€ �s~s�]oooll}}[Zkklbbppqqqqq|css~saaaaazrrrrr^z\r`D```dd=dd ��� �us~s]omlll}}ZZZk}kllnbqqqqqcs]saaaaerrrrrrr^z^D`dបdfπ���� �js~s~olll}kk}l}llpppbqqqqqqqcs~ssaaaaazrrOz8D````dd=dd㞌9g���ƀ �sss~s]o|ooll}}}lbbpqqqq|css~ssaaaaaez⊀rrr٭<`dd=d9Ҍ� �ss~s~ovYoolllnbbpqqqqqqqcs�sssssaaaa^rr`7`D`dd=㞌㌞f�€ �ss~sooollnbppbqqqqooss~saaaaezrrrQ77`ddߌҀ� �~�sjonppqbqqqqss~sssssaaaaarrr9\ۭ`Q7`dd=㞌XX��ƀ�ssBs~sooombbpppbqqqqqccs~sssssaaaarrrrr^{9``DQIdddʞ=X��sss~s�jmppbbqqsss~sssaaarrr^O^````᱖7ddd=㞌VЀ��js~s~sbppbbqqqqqq|css~sssssaae^r{^``````d7ddIŀ� �ss~s~sLppbbqqqqqs~saar{{Q`````d7==+�T�¡ss~�s]_|bbpppppbqq|ccscss~sssssaerrrr{{r``````dI=XX:����T�uss~s]Lo bbbbppppqbbqqqqcss�ssssaaarޙrrrr``````ddddI=ʭ=茌π���js~]sss~pb pppppqqqqqqqoccss~sssssaaaez؀rr`````dឞddIgXXPV���ssjsswmb pppqbqqqcss~ssaaaaaa?zarrr``ۛ```dd㞞ߞ9P9����sssss�sp bpppppqqq|css~ssssaaaazzarrrrrr`````ddd㌛XXP���ss�sc|mpb bppppqqqqqqqcsss~ssssaaaa왨errrrrr``ۀ```dd㞞=XXP���uL~s�sbbpppmpqqqqqccssss?arrr{rrrrr````````dd=X:�ƀ�Yxsss�ssssobbbppppppqqqqccss~ss��aaaaerr{rrrr```ឞdd㞌+Pf�T̀ �xssppbpppqqqqqqqqtsssssss�����aaaaearrrr{rrr``````dd㞌g9X:ŀ��sjsssppppqbqqqqqtjss~w���aaaaarrrrrrr``````dddXߌXXPʀ�E�s!ssooqppbppqpqqqqqqqt~�s]ssssaaaer{rr``ឞ`d㌞ʛьXPXP�€� aLssssqq pppqqqq_L~~~sssaarrrrr{rr9`````d㞞gŌ+:��¡�soӁqqqppbbqbqjomsss]sssaaaaaerrrrr{rr```=````d㞌=9XXg:��a�sssss_oFqqqqqbbbbbbqqqqqqqqt~ss�ssaaaaaarrr^^r``````dXIXX!���ǡC�ssssssqqqqqqbqqqqtt_Lsss]ssaar{r```ۀ``㞌:π ���Ya�ssssqqqqqqqqqbbbqqqqqqqqqqtss�sarr{r```````X:X:H ���a��ssssڀ qqqqqqqqbqqbqqqqqqttttjssssaaaaaarr{````````gXX:Ҁ �w���a�CC�ssssssoYqqqqqqqqqqqqqqqqqqqqqqtqbqt~ssss]ssarrrr{```````㞌=PX:ŀ���aLsssssssuvqqqqqbqqqqqqqqqqqttqqq~sssssaaaaaarrrrrr{``````iX::=��awa�sssssc΋qqqqqqqqqqqqqqqqtttttjssss�arrrrr{``````` 㞌XXPfʀ��wa�sssssY Tqqqqqqqqqqqqqqqqqqqtt_~sss�saaarr{``````XPXf:P�aa�sssssF Οqqqqqqqqqqqqqqt_~sssssssaaaarr{``````=P+:i� �aa�sssssssF qmqqqqqqqqqqqqqqtt]�sssss�aaaarrrrrrrr{```````` XPX̌��aaaa�ssssssoqqqqqqqqqqqqqq~ssssssaarrrr{`````XXX:PP��aaaaasssssF|mqqqqqqqqqqqt_~sss�sa꿊rrrrO````````` 9PPg::̀��aaa�a�ssssc_xqqqqqqqssss�aaaarrrr^``````` XP:X:P��aaaaa�ssssssssc؍qqqqqqm~sssssssaarr{````` X1::̌��aassssssscٍmqq _s~ss�aaÊrrrrrrrr9ۭ````````όXP:gP���awaLaa�sssssscVף _sssssss�aa把rrrۭ```````` gXgPfP1� �?awaasss 'us~ssaaaЀrrr˭rO````` :$̛ � ̀�aaƙsssssss|5 +osss~ssssssaaaaaaаrrr^`````` P1XXP:Pʀ �̀�eبa�sssscu֠ &t_sss~sssss�aaaaЊ⊊rrr˭rۭ`````fI: X�aB�րsssssvàc &|ssssssss�aa rrrr``ۀ```<1f::ŀ�xasssssuG  Xssss~sssssaaa rrrˊrrr˭Q``````8dPfP+XҀ�zFw#asssssscF (덍ssss�aa rrrrrrr9`{`````ۀ"I܌+ f�vBaYssssuF (mssssssaaaa rrr˭rr9````=```fҀPH:&π�ezFaasssFcuY X~j~~~~�ss�ssssaaa rrrrrQ``=`````H9Ҍf !�°aءsss~F|! &Пo|o ~~]L��sa rr9``````f#"P# �͊ze xsj~FcjT &oj  s�s�BCB3Caarˊrrrrrr9```````Ѐ!�I! )�2zewFxssӀ�Fusj&Ѐ�_j| sssssssssCCaw^rrr9`````ឞ  ) � �zϡYss �Fv] 堟Шjo m ssssssaaBazawrrrrrr99```````dО:$!.* ��BzeGG桼~Ys�j oF ssssssssss�Ba2aޙGrrrrr9```Q9``О &P+' � ��Czeea!xYvT~sjTj]cШcFusssssss222a2aew7rrr˭Qr```d P# + f�呂��Ɗz슊aae мásTsjxcF~_Ccssssss�a2B22^2aaarw=r{{r99`````О :$&g����°Czaa$e#!F�sYsjLBcB jcc sRRsssssssssss�22\?2aaaw =r^{{{``X9$ /f&&�������r?a Lv��vsLx~oLLsssssssssBa2a?\2^rrG(=rrr{{{```7``!X+# PPi'S�����eza x?fмL!��]s]T<c7�Rcsssusssssssssaaa2\82aarˊ=rrQ9=```ᔔ`!XI I&):X:gSʀ����€�xzC 7aؓBT�~s Y7cc_]ccccssssssssCaaaB2\22^^r�Xrr9{```۞۞I+ PX:XPgPҀ����rze \axawUT�s ss<7Ycc~_jccssssssssss aaaaO2aa€r�Xrrr9{D````fҌ#.ܛX:Sπ�w�°zŊeaaBwС�ss sLدccT~RRc|vc~ccsssssssssaaaCaa^^aerr�ʭ9``9`=f%f"X%ȌX:PS�T�^?e%aaax��F C~įcc~|jccuvcjccssss�sssUssssaaC^aBzer�XrrrQ­`Hgɀf #/܌PX:X1�ƀ�3HaaewsT &sj]ssFc~c]cuKccK~ssssssssssaaaaa3rrr�99<›````7I*X1P:1:+ ���rzraaaCs &s!~sv3c~cu_csYsjssssCssss aaaazaw7rr"X99!<````X=d!dX'.P:PX,0Ā ��rra aGs &�"ss~s�c]ccsss~s~sssssssY�sssaaaaaaaaa?eLVrr�(r؀``G``۞=!!"&&PPP1PS �€ �^ޓrrza?aaU"GġT &�~sssssF�sss~sssYcsYss��sY�sjzaaaaaaaVrrrrrrr =``` =ʞ 9܌PPX:i � �xrxrree?a؊aKġ &ssss~s�sss]ssFsssFss]sssYaaazъrˀr ŭ^9``T`8Þf!=g# ՌPP::,gSŀ� �rrrrzaаaaaaY" L sw�s~~xss~sAsssssssss�7{aaaaa%rrrU"V^``` !"=ɌfgPP! 9P܌PP1XSS� �rzzr⊰eaaaÿTfj Vswswss~LsssY�~svjsjssssssssssFsTVIaaz\rrrˀr{f8Q`ɔ=ОIX P܌PP:PS,P��CrUrrzeeawaĿL 7wFj~Ywssu�sFYsssssssssFxҡaۀaaaw rrxUQ8`8`8XVI+=X&PPPi:PX+g��GUrrrzwL<B 7BssvGss~sss~ ss �ءwa +aaarw rrrrrU<`7`H7=XXX&PPPPXSgSЀ�™ �rO{rrrrraeaaB١aL ١YTΡuvssǀ�]~sLss�Lء aaaUaaaaaaarw rQUUЛ`Ϟf+=!!X+X&PPPiP:XgSS ���rrrrrrrzeazaaaL 7xvLsuvGsv�Ys!u�<C aw&aarrG QrrЭr3 `8!V=!VЛg&܌PP:XP&+ ��Gr؀rrzea3aBُaj x Yvs€�TLШFT١ aaw'aaaa\\rrr{rrrr­Û`fVI==ў= V&PP.i:PSπ ��͊^rrrrrxaaVaaa 7fLsvjv~ԡj "١a#Йaaa&aaa⊊r{rz!=XI=> Pf&PPPXPii&P �T����rـ rޙaeezxُa 7z %Lvv~jTÿY CFa!7aaa arr{Ur9<`Tz8f" gX Xggg $ X f 'PP.&SSҀ ���T{rr^^axzaaa љ *aB~B!f f FfaaaЊazrrrrTwr{f`9 9=g Hf &.P:XP:P,XS! ���Ozrrr\^eOa %z aC!TB $ ١fa�awaaaހrrrrrr3GQ7{UT{ X Xg= Pf XP1Pi.S ���֭{rrraeaV IaaaYwwLaf$ ١  Ϗa!rr˙r!rrr z<`#8 &Ҟ>V&iPI PSgSSЀ��{\r{rrrrrzފa"7 ŏaaCxxϧ�$f HafwЊaBC^ arrzr7 x%`!H8 >&/ :$PP:PS1��xzrrrrzrrrrފrO   V&aaaxďwءawT H ďaT$aw<C!exGzTrz <`xHf78(>>> &1܌(1XSSSр��{rrrr{rrrrO 7&ŏafaBwGпaxGwffϨ ШCzUHrr z *Û7"H* (->! (PP:PPP,SSS��7rr˔rrrzr?eU V&aaaaaa�Taw a3�ffϏfe#feGGGrrOح^ `U`7""* & ,>>#&9܌܌P1P1SgP ��Tz{rOr?reCz! &aaa7aa O axxa$f"aafeewGUrUíQ U`7"! ' & *ߞf(1PPP P1S! ��zQ\r{rrrr Й#$ aaaޏaa؏a faaYxЏ "O e OGUа"^{ ҭ`7` , &->>f(܌܌܌܌P PP:SgPS+ �T�ƭ{{rrrrrrrĊOЙ! ۊaa aBa 'a VaCxؙ"2  aΊwr3rz3 H < ``H{` -  ʞ>1g܀PPPH P,S ��T�zQ{rrrr^\rrr reO7z Ű a ŏa Vaaz؏fzCBЊ rrrBf"rHx"ЭwH `Uf`О -g(' +>1 (܌PPPi&P:SSg �ƀ��T{{rrrrrrrr$rrG e e찰a ۏ ޏaؙzЊ^ Brrr Q Ix ѭ7= `9``U`Iў!(-&(&,>>1 (܀PPPP PPP:,PX:S!���7r{rr{rrr$rrG^" ŀ aHG \Hޏaxèzefzxr{rr^{rЭ &{ {z (```U``"`(>>>ߞ1 (܌܌PPP:i9PPPP,i*10���`z{rrrrr^frrrGr ۰eѰ ްHaeeؙCGUUа{ rrrr^ r Xz{ Q7 `````UI`Hf>(>>>>1 9ߌ܌PPP:.P:P,X0��€�ح8O9rrrr^rrrrGrr ŀUI^ee ^eeaeaz#wwаz 'rrr^^r &rgȭxrV7 -`````8``gĞ(>>>1 ܌PPP1ՀP+0:Ҁ��Tƀ�Q{rrrrrfrrrr rw{eUe zeeبfr ۀrrr XrXЀQ ``бU``H`&9>1rz ʞPPPPPXP:PP: &P1 ��� �w`z rrO{rrrrrwrr grxarЀ&eeee氊rO{frV +rrzЭrr&Xrz -`+`&>>ߞ>1^ ឌPPXP,:SЀ �€$� ޖ< r\Orrrr! grrrr?e Tаee^Of< rrrz rr!XrOO`` X`<  &ឞ&>ߞ1^C^PP/iPP/P:0-P, �$�ƭHD {r\rrrrCrrr! +rrаrrrr*rrrrr rQ {rf ŀr3 rrH )Xr{`` X`&X= >>1߭V^P܌.P:PPi PS# ���#�ح7` ޭ^^rr\rrr =rfrrrrr^Irrrrrrzrr&rIrr +rT r ^^O```&X``9X*ߞ111\?IQP19P:PPP01S ���v€$�^\`D O⭊^rrfrrT rr7rrr?rrrrrrrrrr?rrr rr'rQ+{`ɭ````&&```ឞ8!99d+ߞ111ߞۭ7Q>/P&PSX&P���w%�GHDr\\rrrfrr!=rrrrrrrrrrrފrrrCrrrrr =rrT&%{ۭ````='X````8(՞Xdd111{```PP&:P:PP$PP"��w��%�<H```DO\QrrT rrrrr?rrrzrrrrrrr&rr!(r˖``````=X```` =܌g"PPXPP/P:P0P%1/��'�^ɔ```DOޭrOT rrrrrrrr\rrrrrrrrrrrrrrrr(rr& `ۭ````ឞ= 9X1܌ČPP& P'-����(�GQ`D`r\O!rrrT rrrrOrrrrrOrrrrrr^r!r{&rrr  X``ۭ````````````` d 1܌XPP$Pܞ,1P/.%/ ����)�x````D`D`ހOw r⭭rrrrrOrrrOrrrrέr&؀{ ({U '{{{^Q````ឞ```9XdP܌܌܌XPPP->P/f1/f �T4�{Q<```D\ rrrrrrr^^rr♭rrrQ{OOU{{&zr &``fO۔Э`````````X=1܀XPPP,iP/1%+ ���€0�˔HD`D``DO rrrrrr?QrOrrQO\rr&zz X```````ۖ۔ۭ````fI1܀P߀P1P/i'X.΀��0�G7D```D`DDr X⭭rrrrrrOQOOO\⭭rrr"(zT& X`````````=g=` XX1܌܌܌XPPP91P/.+/"�0�x``ᭊ```\rTX rOrzOO\⭭r x&\^T'` X````ឞពg9螌PP܌܀PPPX:PPXiP:P/1 ��T1�7Q`D`D``D`D`r  OOޓ(z'```&````````````=9"XP1܌܌܌9ߌ:X P/X..�T��2�`DD``D``Q X{zrOrr 3(8!'``(``=``````==ʛߌ܌1PPPXP1X P1X.X/X����3�ɔ`D`D``D` QOO\rހ(z`+```````````=``===ҞP܌܌܌PPPX1ʀP:P,/.��5�ƛɖQDDD```&`DD Q?{O  &Q`````````=```` >gXPP܌PPPXʀP.,P/X1!��ƀ7�df=````D<&```DDDrހD  +`8``7``7 g```````ߌ=>P܌܌PPPP.P1XiPP/X/"�������:�UXdۭ`````&D`D``DDDr7DDDD`D`D! X```'``````۞`i999P܌܌PPPPśPP̀P1X-P0P//Ѐ���w<�ĭۭ````<(D```DrD``D`7Q`D``D`! X`z```7 =```````ɀ>1iXiP܌PPPXXPPP:P1X-:P0PP+1��=�H=Q٭``D`(Q\D᭱`````D```Q``D``D``` X````{ X````````g`ឞ=>19Ȍ܌܌PPPP1PP1X-P:01 �F>�H{ʖdfDD`᭔`D`QD````D`Q\```D᭱D  &D`8`` ```````= >9܌9PPPX.PPPiPPP:0P01/ ���?�H=ʔd`+D``Qᖭ`````D``D``DDᔱ=`D`D``D``D X``` ``````ۀɀ i1܌Ռ܌PPP1XP1i:P:P0PS1P ����ƀ@�Hd`dX`Q````ᱱ``\`D XDɱ``f``````== i1܌P܌țPPPPPP,SPP0PPS%10f��B�xdបX`D۔```D``D``D``DD````D``D``` X`97 `````d111P܌PP91P+1P.g:SP0:P011S ��C�UdʔdQd7I9បᖖ``D```T (ឱ{dV`ឞ`=d=ț܀܌PPPXPiiXSSPSS0Pi$1S��G�{ddd7˔7`ۭ``D``D``D``=`D```d 9`ɞId`Ȁ9܌܌܌܌܌PP.+X.::ii00P0P���N�9ٛd=7X=Xdឱបឱᱞd` 9d{dddd==XiP܌PPP9XPP0SS0gS.#���N�=XddI d=QdddDDពឱ ddddddX۞>&PPP1܌܌ՀXPX:P1SPSSPi0,���N�z==d{d dgdឱdd=7<Dឱ=ddd9ឱ< 9dd۞dddddd=X9܌ȀXP.1P:X10SSXiS+���O�dI dddd7=Xdddᱞddd&9dd{dddddddʌŞPPȌ܌܌ՀXȌPP1XP:PSSXiS��������N�ۛdd dd=ddddddd=dddd7՞ddddd=㌌gX>1X=XߌP:P1P::PSSPSPg ��w�������O�T====dd=ddddd=Xdd7====^ddddddd7㞞dd>g9fXX=XXX9PP1:P:P:PSii0iS! ���������O�===dddddddddd=ddd=={=V! ۀ=d㞌㞌=V==X=X==X=XVX9 PPP1PPP̞10SiXS �w��w������Q�ś=g=dddddddddʞdddd= d=р =ߞ=߀ PP:PP:::P:SX1 ������w�T�{===V㞞ddddddd==d=ddddddI㞌gI9 PPPPXiP:P:P0SSgS ������w�\�UI㞛V=d㞞㞞㞞==ddd=㞌Ō㞌=XۀPPPX::P:PPgSՌ1Xi ������_�=9ʌ==ʌ㞞dddd=Idddd㞌dŌg+PPPX:::PP:PSS&1SS+ �ƀ�Fa�X==㞌===ʌ㞌㞌=㞌㌌g1=9PPP1:PP:P::PSPS1$XSi.��T���c�fیߌ==㞌㞌===ߞ==߀Xi+PPXP:P::PP0S��v���f�=ěi===㌌==d=ߌҌi9=߀+PPXX.PP::P:P0iigS+���m�GIV====ʌ=9g=9=ۀ==PPXX̌̌:PP:PPSiiiP"���Tn�gfXʌ===㌞ʌ=X=ʀX9Ō9ߌ=ȀPiVgP̌:PP:::P,iSS&P���q�UgIIXX=X=ʌ= ===9=9IXXiiX++:X̌̌PP:P:PP,,,S1+i�s�==ћ=XIX=Ȁ i=X=+g++XIX+++̌P::::::P::1XSSPSS P�v� fʀX=g=g==gX=XXXffVXʌ̌P̌P:P:XXXȀ0SSŌX"!�y�!IIIՀ9XXX蛌1XXXXXXXX:PS0SS!$�ƀ~�=Xi i XXX9PX:X:̌:::X::::PP:0S0&Xif'��IgfXXՀXX9P1:::XXǐP1:P:P:PPSSPSgXՌ*)�€�"I=Xۀ XрXPPPXPP::iXXXP:PPSSg+�Ig+ۛʀPPP:PP:::PP::XX,, 1+�XPP1XPPPPPP::PP:P::X:SV#�7fffǏ̌P̌1PP:::1P::P:Pi,$f$I�!fgfXXX.̌:P::P:1XP:g,$$$Xf�Vfg$f$ffXXXXX$$X9!�f+ț1P:P1igIf� =EndInset ll��7����A� {:��ll��7���� ����l��� ll��7����B�{����ll��7�� ������������Wz��` ��d������6�������� �������� HH��ˆ�����o�����HH��ˆ��������������WvUTUT��a HH��ˆ����6����HH��ˆ����  ����l��� HUa������o�����HUa����D�����������W|UTUT��hV�LINSTITUTE FOR SIGNAL AND INFORMATION PROCESSING�MFebruary 15$, 1999 HUa�����6� ��HUa�����  ����l��� H$������o�����H$����������� ����W}UTUT��hG�NImproved Monosyllabic Word Modeling on SWITCHBOARD�O H$�����6����H$�����  ����l�����d������:#�������� �������� l ���������o� ����l �������������� ����~�e"<$paranum><$paratext><$pagenum>  �e"<$paranum><$paratext><$pagenum> Q�e"<$paranum><$paratext><$pagenum> l �������� B� ���l ��������  ����l��� l GH��:%� ���l GHl l �Table of Contents Specification��d������:&���������������� l ����u���o�����l ����uH������������ ��e Level2IX ��e Level1IX ��e 1, 23 %��e$<$symbols><$numerics><$alphabetics> !:��eLSymbols[\ ];Numerics[0];A;B;C;D;E;F;G;H;I;J;K;L;M;N;O;P;Q;R;S;T;U;V;W;X;Y;Z S E��e <$pagenum> l ����u�� C����l ����u��  ����l��� l1Km0��:(����l1Km0ll�Index Specification��d������:)���������������� l ���������o�����l ������ ������������W��e"<$paranum><$paratext><$pagenum> l �������� D����l �������� ����l��� l G,��:+����l G,l l �List of Figures Specification��d������:,���������������� l ���������o�����l ������ ������������W�e"<$paranum><$paratext><$pagenum> l �������� E����l �������� ����l��� l GK��:.����l GKl l �List of Tables Specification��d������ I�������� �������� $$�������o�����$$��������.�����.$).38=Bx{~PTX\`dhlpt/258;>ADGJM#'+  �W��Bm������ $$������ L�����$$�����  ����l�����d������ �������� �������� $$�������o�����$$�����������C �$).38=Bx{~PTX\`dhlpt/258;>ADGJM#'+  ��$$������ �����$$����# ����l�����d������ ��������## �������� $$�������o�!����$$����J��������D"#�DGJM#'+  ���$$������ �!����$$���� �"" ����l�����d������ g��������&& �������� $$�������o�$����$$����������Gm&����W��Bm� $$������ j�$����$$������%% ����l�����d������74��������(3 �������� HmR��75�'�)��HmRHRHR�Footnote Hr@��76�'(*��Hr@HzHz� Single LineH�����77�'),����++��������Footnote � �����78� *���� � ���� � ��� HD��79�'*-��HDHH� Double LineH�����7:�',0����./�������� Double Line�������7;� -�/��������������������7<� -.��������������H�����7=�'-2����11�������� Single Line�������7>� 0���������������HZ����7?�'03�������������� TableFootnote EGxR��7@�'2���EGxREPwEPw� TableFootnote��d������:/��������66�������� ll��7�����o�4����ll��7���L�����6����9���` :���` !{��� 'Improved Monosyllabic Word Modeling on ���@ SWITCHBOARD ;���` C���` E���` 0���` 1���`February 15, 1999 F���` .���` $���` G�� h�" HA���` IO���` J���`submitted by: K���` L���`DV. Mantha, J. Hamaker, N. Deshmukh, A. Ganapathiraju, and J.Picone M���`0Institute for Signal and Information Processing N���`2Department of Electrical and Computer Engineering O���`Mississippi State University P���` Box 9571 Q���`413 Simrall, Hardy Road R���`%Mississippi State, Mississippi 39762 S���`Tel: 601-325-3149 T���`Fax: 601-325-3149 AU���`2email:{mantha, hamaker, picone}@isip.msstate.edu ll��7����+)�4����ll��7����55 ����l�����d������*��������99�������� HH��ˆ�����o�7����HH��ˆ���D)���)����9�������`EXECUTIVE SUMMARY  $��� USWITCHBOARD(SWB) Corpus consists of 2438 conversations digitally recorded over long 02����^distance telephone lines. The SWB Corpus totals over 240conversation hours (elapsed time) of ����]data. The average conversation duration is six minutes. The transcriptions contain more than ����^3million words of text. The SWB Corpus includes more than 500adult-aged speakers and covers ����Vmost major American English dialects. Such impressive statistics make SWB the premier ����\database for telephone bandwidth large vocabulary conversational speech recognition(LVCSR) ����eresearch. The goal of this project is to resegment the speech data and correct the transcriptions in ���@5an effort to significantly advance LVCSR technology.  ����� cThis project has reached it)s midpoint. Thus far, we have released 367 conversations with 0������Xcorrected segmentations and transcriptions, and another 551 conversations with complete ����Wsegmentations. These conversations comprise 50% of the conversations used in the WS97 ����^partition and 40% of the entire SWB corpus. We have also demonstrated that one could obtain a ����W2% decrease in WER by simply reestimating LVCSR models on the corrected segmentations. ����VThis work was presented at the recent Hub-5 Conversational Speech Recognition (LVCSR) ����\Workshop and was met with much support as well as many suggestions for improvement from the ����Tspeech research community. Keeping these suggestions in mind, we have made numerous ���@Jrevisions to our segmentation and transcription procedures. These include  #&��� Zminimizing the amount of nonspeech data included in an utterance definition. We are also  2���@qattempting t#o maintain as much phrase structure as possible while still allowing for silence paddings;  D��� oa #new workflow process in which we do  segmentations f#irst, and then provide transcriptions in a #P����psecond pass through the data. We plan on completing the segmentations by March #31,1999. To this ���@?end all our validators are presently working on segmentations. q��� dWe have also implemented an incremental and multiple)pass quality control procedure which 0����aprovides almost immediate feedback to the validators. For a database of the magnitude of SWB, it ����Yis imperative that the quality checks be done almost immediately after the data has been ����[segmented. This approach has led to a decrease in the error rate as well as an increase in ����Uproductivity. We had earlier reported a cross-validation of less than 1% WER for the ����btranscriptions of a relatively clean utterance. These figures show a substantial improvement over ���@Xthe current LDC transcriptions which have an 8% WER measured under the same conditions. ��� �By March )31, we will release the final segmentations for all )2438 )Switchboard conversations). 2)����dThis will be a major milestone in the project, and be a sign the end is in sight. We then will turn ����dour attention to transcription correction, and project a completion date of August31,1999. Such a ����etimeframe will allow us to complete the manual and automatic word alignment) generation and ����hreview before the projected deadline of December 1999. All information relevant to this project is ����nalso available at" http://www.isip.msstate.edu/projects/switchboard/), including the most current set )���@of segmentations. ,Q��� cOur Switchboard mailing list has recently grown to 23 users, an indication of the renewed interest p_����ain this project and its data. Discussions are underway to support several projects interested in ����cusing the data for tasks ranging from phonetic recognition to prosodic labeling. Reconciliation of ���@Kthe data with other SWB resources is becoming increasingly more important. HH��ˆ����*�7����HH��ˆ����88 ����l���eBۈ���C�!{���eBۈ���;; ����l��� eBۈ���D�{����eBۈ���������:����W UTUT���h:� INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING�!  D��%���������B����D��%�������D�����H2>������@h� W!���h'�An overview of the released data. D��%���������BGOOG��D��%��������==����l�����d������&��������AA�������� HH��ˆ�����o�?����HH��ˆ��‚���D(���(����A�������h�#ABSTRAC T W $���(uIn our last report[�1�]  we described an improved workflow process based on a new set of 02����Zsegmentation and transcription guidelines. These new guidelines were a result of numerous ����lsuggestions received from colleagues during the initial months o f this project. We a lso "����ii mplemented an incremental and multiple pass quality control procedure. Cross-validation tests �����s howed that these new procedures decreased our W ER f rom 3% to less than 1%  (the original LDC  ���@Otranscriptions had a W ER of 8% measured under the same conditions). ����� ~In this quarter we h ave not made any c hanges t o the segmentation and transcription conventions. 0������fWe did, however, m ake a few adjustments to our work procedures to counter abrupt changes in ����dstaffing. Key among these is our decision to only segment data (no transcriptions are changed) "����lu ntil the entire corpus has been completely  segmented. This new procedure gives our new ����xvalidators the chance to be productive almost immediately. W e plan to complete s egmentation o f  ����rthe entire database by March )31 a nd then shift our focus to transcriptions. Though we are slightly ���@jbehind schedule, the projected completion date of this work remains at D ecember1 999. y�����h� INTRODUCTION  ���(The SWITCHBOARD Corpus[� 2� , �3�-] has become critical to the success of state-of-the-art LVCSR 0 �� ����`systems. Using this data, however, has not been without its share of drawbacks. SWB was a great ����pexample of the trials and tribulations of database work in that the quality of the data suffered from ����Za lack of understanding of the problem. Word-level transcription of SWB is difficult, and ����_conventions associated with such transcriptions are highly controversial and often application ����^dependent. By 1998, the quality of the SWB transcriptions for LVCSR was recognized to be less ����`than ideal, and many years of small projects attempting to correct the transcriptions had taken ����\their toll. Numerous versions of the SWB Corpus were floating around; few of these improved ����atranscriptions were folded back into the LDC release; and many sites had spent a lot of research ����dtime cleaning up a portion of the data in isolation. In February of 1998, ISIP started a project to ����acleanup the SWB Corpus, and to organize and integrate all existing resources related to the data ���@into this final release.  ����� eIn the first six m onths of this project, we made considerable progress in transcribing and 0�� ����Uresegmenting the corpus. We released 1000 of the 2438 SWB conversations with revised ����bsegmentations and transcriptions. As noted in our last report, though, we r evised those ���zconversations to conform to some very important conventions �)[4]�( (marking of boundaries near ����_noise for example). We also amassed a large collection of tools and resources for use with the ����YSWB project. Most notable of these are the development of our public-domain segmentation ���tool �[5]�, the SWB frequently asked questions (FAQ) web-site�*[6] �, the SWB educational resources ���web-site�+[7] �, and a comprehensive collection of statistics �,[8]� related to SWB. We continue to ����gmaintain a mailing list ( swb@isip.msstate.edu ) which is our point of contact to the research ���@`community for resolving subtle transcription issues and communicating progress on our efforts. (b����� eOur plan at that point h ad been to continue working on segmentation and transcription p������nsimultaneously, finishing the S WB Corpus by the beginning of August1 999. However, in the last B ����fthree months of this work we have been forced to alter o ur previous plans. This was done to HH��ˆ����)�?����HH��ˆ���D@@ ����l�����d������D��������DO�������� HH��ˆ�����o�B����HH��ˆ��‚���D'���'����D����( ����oaccount for changes in staffing and to make)  up for the time spent in training of the new validators. ���@`These changes and the adjustments made in our schedule are described in the next sections.  2���h&� CHANGES IN WORKFLOW PROCESS ��� ZTo insure that the released data is in accordance with the transcription and segmentation 0\����bconventions, it is important that the validators work be reviewed almost immediately after it is ����^completed. This provides the validator with immediate feedback and allows the project manager ����ito prevent any long-term problems in the data. In the previous quarterly report we described a ����\process wherein some validators were working on segmentations and others were focusing only ���@con transcriptions. However , staffing changes have forced us to reconsider this strategy. \����� gTo get our new validators up to speed quickly, we have decided to allow them to focus on the ������srelatively simple task of segmentation. It is our plan to  first complete segmentation  of the entire  ����{corpus and then to  again  churn out revised transcriptions. A majority of the effort spent on this "����^work is in transcription and word alignment review so the amount of time required to complete ����gsegmentation of the corpus is minimal . Yet it is sufficient to allow our newer validators to gain  ���@,experience before beginning transcriptions. ]����� uLimiting our validators to only segmentations provides two other advantages . Fi rst, the research 0 ������[community will have the opportunity to begin working with the new segmentations across the ����\entire corpus. It is our opinion that this will better serve the community than waiting for ����cincremental releases of corrected segmentations and revised transcriptions. The other advantage is ����kthat this lets our new validators be productive almost immediately. The  average training time is ���@]less than one week for segmentation; training time for transcriptions is nearly triple that. xt�����h�3RELEASED DATA ` ��� _Since our last progress report,  we have released 162 more conversations with revised 0������]transcriptions and 552 having only revised segmentations. We have also released the complete ����iSwitchboard database (with a mixture of LDC/ISIP transcriptions) to satisfy r equests from ����xmembers of the research community.  In total, we have released 362 conversations with completed ����fsegmentations and transcriptions and another 556 conversations with complete segmentations. "����/This comprises 151 hours of conversation data. ����5Another 250 conversations  will be released  ����+shortly. The released data covers the ����ZWS97devtest set, WS97eval set and part of the ���TWS97training set.  Table�1�  gives some pertinent ���@*details regarding the released data. )8����� 1We will continue to make incremental releases of pF������&newly segmented data available to the ����.community via our project web page. A revised ���Ftimeline for these releases is shown inFigure�$1�. ����0We will make a release of the entire SWB Corpus ����2with revised segmentations on March31. HH��ˆ����G�B�G��HH��ˆ��ACC ����l�����d��������������|��������uL������+E��uL����u$uL;��%���������-BD>O�>�;��%�������}JUS1�����B�K��JUS1��=�[�W���`Conversations ��d������w��������UU�������� HH��ˆ����x�I����HH��ˆ��X���D!���!����U����| ���h�7ACKNOWLEDGMENTS %$��� cWe would like to express our sincere gratitude to our sponsor, the Department of Defense for their 2����kunwavering support  of this project.  We also appreciate the assistance of Dr. David Graff of  ���@TtheLDC for his help  in various aspects regarding the S WB data. &\���h�REFERENCES ��� VJ. Hamaker, N. Deshmukh, A. Ganapathiraju, and J. Picone, Improved Monosyllabic Word �����HBModeling,  Department of Defense, August 15, 1998.�  "�����([�.J.Godfrey, E.Holliman and J.McDaniel, Telephone Speech Corpus for Research and De������\velopment,  Proceedings of the International Conference on Acoustics, Speech, and Sig-  ���@Onal Processing, pp.517-520, San Francisco, California, USA, March 1992. *�����(Z�%B.Wheatley, G.Doddington, C.Hemphill, J.Godfrey, E.C.Holliman, J.McDaniel, and ������OD.Fisher, SWITCHBOARD:AUsersManual,  http://www.cis.upenn.edu/~ldc/  ����^readme_files/switchbrd.readme.html, Linguistic Data Consortium, University of Pennsylva- ���@nia, December1995. ' �����(_�'J.Hamaker, Y.Zeng, and J.Picone, Rules and Guidelines for Transcription and Segmenta- 0.������Ption of the SWITCHBOARD Large Vocabulary Conversational Speech Recognition Cor- ����]pus,  http://www.isip.msstate.edu/resources/technology/projects/current/switchboard/doc/ ����btranscription_guidelines, Institute for Signal and Information Processing, Mississippi State ���@University, July 1998. t�����(\�&N.Deshmukh, A.Ganapathiraju, R.Duncan, and J.Picone, An Efficient Tool For Reseg- �����^mentation and Transcription of TwoChannel Conversational Speech,  http://www.isip.mss-  ����nt ate.edu / resources/technology/software/1998/swb_segmenter, Institute for Signal and Infor���@>mation Processing, Mississippi State University, August1998. ����(\�J.Hamaker and J.Picone, The SWITCHBOARD Frequently Asked Questions(FAQ),  ht ������]tp://www.isip.msstate.edu/resources/technology/projects/current/switchboard/faq, Insti- ���@Wtute for Signal and Information Processing, Mississippi State University, August1998. �����(W�J.Hamaker, A. Ganapathiraju, and J.Picone, SWITCHBOARD Educational Resources, 2 �������Yhttp://www.isip.msstate.edu/resources/technology/projects/current/switchboard/doc/educa- ����ation, Institute for Signal and Information Processing, Mississippi State University, August ���@1998. 8����(c�J.Hamaker and J.Picone, A Statistical Guide to SWITCHBOARD: Topic Statistics,  http:/ F������`/www.isip.msstate.edu/resources/technology/projects/current/switchboard/doc/statistics, InB���@Zstitute for Signal and Information Processing, Mississippi State University, August 1998. }T1��X���BHL��T1��X=�[W���`918 }JUSD���BKM��JUSD=�\�W���`# of non_silence utterances }TDX���BLN��TDX=�\W���`71948 }JUSX���BMP��JUSX=�]�W���`# of silence-only utterances ;��%���������B>���;��%������G>}TXX�� �BNQ��TXX=�]���` Q���`40452 }JUSx�� �BPR��JUSx=�^�W���`hours of data }TxX���BQ/��TxX=�^W���`151.42 Hz����V�����E����Hz����V��@��������������)����]Following this, we will begin to make incremental releases of the revised transcriptions. We 0����_expect the delay incurred in switching the new validators back to transcriptions to be minimal ����nsince each will have at least two months of experience at that point. A major release of the corpus ����Ywith revised transcriptions and segmentations is set for early June. This is followed in ���@ZDecember1999 with a final release of the entire corpus including manual word alignments. a\���h�ISSUES t ��� ~As this work continues it is clear that each validator reaches a  point where they simply burnout . In ������ithis quarter we lost our two most experienced validators for this reason.  Each had been with us for " ����dover six months and neither returned to work after the Christmas break. Of course, this practically ����]stalled the project until we were able to hire and train a number of new validators. We fell ����^approximately one month behind schedule during this period of training. The delay was limited ����wsince we were able to get the new validators i nto p roduction mode quickly by allowing them to ����bonly work on segmentations. Our most successful validators associated with this project have "����_survived about ninemonths. We have good reason to believe that one year is the upper limit of ���@6what we can expect from a good hourly student worker. ^ ����� ~Though  these  setbacks were  devastating at the time, it appears that the work is getting back on 0������_track. Our current set of validators are actually producing more data per week than the former ����]validators. Our initial examinations of the new data also shows that the quality of the b����osegmentations has not decreased with the new workers.  If this trend continues, by late March  we ���@<will again be on track for a December1999 completion date. HH��ˆ����z�I����HH��ˆ���JJ ����l���*SuL������+E[��SuL����Su$SuL H�9���+EZ\�� H�9� H)KJKy H�9���+E[]��Ky H�9�Ky HKy)KJ| H�9���+E\^��| H�9�| H|)KJ H�9���+E]_�� H�9� H)KJ9P H�9���+E^`��9P H�9�9P H9P)KJY# H�9���+E_a��Y# H�9�Y# HY#)KJ H�9���+E`b�� H�9� H )KJP H�9���+Eac��P H�9�P HP)KJWv H�9���+Ebd��Wv H�9�Wv HWv)KJ4;-����+Ece��4;-��4;-֗;-4.J����+Edf��4.J��4.J֗.J4!i����+Eeg��4!i��4!i֗!i4����+Efh��4��4֗4q����+Egi��4q��4q֗q4Kˆ����+Ehn��4Kˆ��4K֗KSuLg�����z�)Ek��SuLg���Su6ݨSuou<`AuLSuL��9��{�*Ejl��SuL�NSSu Ӛ ".Kur&+yuLSuLZ� ��|� Ekm��SuLZ���Su ]jsSt3D(z#uL u\/},?��}�+El��u\/},?uc]uc]�* segmentation ,G?���+Eio��,G?33�*MAR G,⓰?���+Enp��G,⓰?G3G3�*APR ,ˉ?���+Eoq��,ˉ?33�*MAY 0?A,⒝?���+Epr��0?A,⒝?0?A30?A3�*JUN P,?���+Eqs��P,?P3P3�*JUL nĄ,=?���+Ert��nĄ,=?nĄ3QnĄ3Q�,AUG B,)?���+Esu��B,)?B3B3�*SEP 8,7h?���+Etv��8,7h?8383�*OCT t/,F?���+Euw��t/,F?t/3t/3�*NOV mA,7h?���+Evx��mA,7h?3mA3*DECSu$Z�����+Ewy��Su$Z���Su$u$ H�9���+Exz�� H�9� H)KJx�$k���+Ey{��x�$k�xx)3 ,⒠o?���.Ez|��,⒠o?33�*FEB `sWT?���E{}��`sWT? k`k*400 `3T?���E|~��`3T? a`a*800 +&Lp?���E}��+&Lp? T+T*1200 +ŕLp?���E~���+ŕLp? G+G*1600 +m Lp?���E��+m Lp? t:+t:*2000 ,1HL�?���E���,1HL�?1O.,1O.-2400Hz����V�����E��Hz����V��DUSS����l��� M^Ǵ3މ?��~�+Em��M^Ǵ3މ?MeMe�* transcription Su^Ee?���+E|��Su^Ee?SueSue�*word alignments ,ʑ?���.EF��,ʑ?4 4 �*JAN` �9���+Ej��` �9�` `)M g�w?\���v�E��n̯\?nn�'No. of Conversations >y?ƖN?��y�EF��>y?ƖN?>yG >yG �'1999}JUSŒ���BR0��JUSŒ=�_�W���`hours of speech data }TŒX���B/1��TŒX=�_W���`93.09 }JUS ���B02��JUS =�`�W���`Mean utterance duration }T X���B1���T X=�`W ���`4.66 THQ�����t�E�Z��HQ���������l��� HQ�����u�E����HQ���������������#���(c�Timeline for the remainder of the SWB resegmentation project. Our anticipated completion date P���@remains at December1999. ��d����������Left�d���������Right�d�����{��isip_coverpage�d�������� isip_preface�d������� ��TOC�d�������IX�d�������LOF�d����� ��LOT�d����� ��HTML�d����� ��HTML�d����� !��HTML�d����� $��Headings�d����� �'�� Reference�d�������4���isip_coverpaged�����7��� isip_prefaced�����?�����d�����B�����d�����E�����d������I������:����f�@��������g�������� ���������������� ���������isip_appendix_0����Appendix <A+>.�. ��� f�@���������g�������� ���������������� ���������Z�� ff��. ���� isip_appendix_0_toc����APPENDIX <A+>.\t�. ��� f�@���������g�������� ���������������� ����������isip_appendix_1���� <A>.<n+>. �. ̜�f�@���������g�������� ��������������������������� ff��. ���� isip_appendix_1_toc���� <A>.<n+>.\t�. ��f�@��������R�g�������� ��������������������������� isip_attachment���� R:[<n=1>]\t�. ����f�@��������R�g�������� ��������������������������� isip_attachments���� R:[<n+>]\t�. ��� f�@������������������� �������������������������� isip_body������.  �f�@����������g��������� ��������������������������� isip_bullet����\t�. �� �f�@����������g��������� ��������������������������� ���� isip_bullet_text����\t�. BEߑ�f�@���������g�������� �������������������������BE�� isip_enumerate����<n+>.\t�. �f�@��������E �g� ������� �������������������������� isip_equation����E:(<n+>)�. ������@��@��������F���������� �������������������������� isip_figure����F:Figure<n+>.�. ������@��@������������������ ��������������������������isip_figure_part����(<a+>)�. K���f�@������������������� ���������������� ���������K�� K�� isip_figure_text�������������@��@��������������������� ���������������������������Body������. ������@��@��������F����������� �������������������������� isip_figure����F:Figure<n+>.�. ������@��@��������F����������� �������������������������� isip_figure����F:Figure<n+>.�. �����f�@������������������ �������������������������� CellFooting������. ���@��@�������������������� �������������������������� isip_footnote������. ���f�@��������P�g�������� ���������������� ����������isip_heading_0���� P:<n+>.�. ��� f�@���������g�������� ���������������� ����������� ff��. ���� isip_heading_0_toc����<n+>.\t�. ��� f�@��������P�g�������� ���������������� ����������isip_heading_1���� P:<n>.<n+>.�. ̜�f�@���������g�������� ��������������������������� ff��. ���� isip_heading_1_toc���� <n>.<n+>.\t�. ��� f�@�����������g��������� ���������������� ����������isip_heading_2������.  � f�@����������g��������� ���������������������������� l�� isip_indent����\t �. �����f�@���������g�������� ���������������� ���������� isip_misc������. ��f�@��������R�g�������� ��������������������������� isip_reference���� R:[<n+>]\t�. ���f�@�����������g��������� �������������������������� isip_sta