The decoded and reference scripts were filtered using the following rules
before the word error rate was computed.

Also, all instances of spelled out letter were treated as
separate words. For instance, IBM would be treated as I B M
for the purposes of scoring

    $_ =~ s/-/ /g;
    $_ =~ s/!/ /g;
    $_ =~ s/ E MAIL / E-MAIL /g;
    $_ =~ s/ EMAIL / E-MAIL /g;
    $_ =~ s/ E MAILS / E-MAILS /g;
    $_ =~ s/ EMAILS / E-MAILS /g;
    
    $_ =~ s/"WHO'S"/WHO IS/g;
    $_ =~ s/I'M/I AM/g;
    $_ =~ s/NUMBER'S/NUMBER IS/g;
    $_ =~ s/"YOU'RE"/YOU ARE/g;
    $_ =~ s/WANNA/WANT TO/g;
    $_ =~ s/GONNA/GOING TO/g;
    

    $_ =~ s/ WESTPALMBEACH / WEST PALM BEACH /g;
    $_ =~ s/ WESTPALM / WEST PALM /g;
    $_ =~ s/ MAGICKINGDOM / MAGIC KINGDOM /g;
    $_ =~ s/ DISNEYWORLD / DISNEY WORLD /g;
    $_ =~ s/ FEDERALEXPRESS / FEDERAL EXPRESS /g;
    $_ =~ s/ UNITEDSTATES / UNITED STATES /g;
    
    $_ =~ s/ MATCHPOINT / MATCH POINT /g;
    $_ =~ s/ LOTUSNOTES / LOTUS NOTES /g;
    $_ =~ s/ MAILROOM / MAIL ROOM /g;
    $_ =~ s/ NEWYORK / NEW YORK /g;
    $_ =~ s/ BLUEMEADOWCOURT / BLUE MEADOW COURT /g;
    $_ =~ s/ SPEECHHUMANFACTORSGROUP / SPEECH HUMAN FACTORS GROUP /g;
    $_ =~ s/ GIGGAINFORMATIONGROUP / GIGGA INFORMATION GROUP /g;
    $_ =~ s/ SPEECHSCIENCE / SPEECH SCIENCE /g;
    $_ =~ s/ LOTUSFAX / LOTUS FAX /g;

    $_ =~ s/ A P I / A. P. I. /g;
    #$_ =~ s/M\.S\.P./ MSP /g;
    
    $_ =~ s/ ISSCVM / I. S. S. C. V. M. /g;
    $_ =~ s/ MSP / M. S. P. /g;


    $_ =~ s/ SIMPLYSPEAKING / SIMPLY SPEAKING /g;
    $_ =~ s/ VOICETYPE / VOICE TYPE /g;
    
    $_ =~ s/ MISTER / MR\. /g;

    $_ =~ s/" ASKIN'"/ ASKING /g;
    $_ =~ s/" GOIN'"/ GOING /g;
    $_ =~ s/ FIREWALLS / FIRE WALLS /g;

    $_ =~ s/=EM/"'EM"/g;
    $_ =~ s/=CAUSE/"'CAUSE"/g;


The reference script used for scoring is ../eval/scripts/eval_scripts_filtered.all.