source: liacs/mms/assignment2/wer.pl@ 405

#!/usr/bin/env perl -w
#
# Little algoritm to determine the Word Error Rate of speech recognition
# systems detection.
#
# Rick van der Zwet <info@rickvanderzwet.nl>
# Licence: BSDlike - http://rickvanderzwet.nl/license


# While processing look this many words ahead for potential matches. Roughly
# spoken this number will limit your ability to detect at maximum LOOKFORWARD
# number of insertions and substitution in one detection cycle.
$LOOKFORWARD = 10;

# Correct user if needed.
if ($#ARGV < 1) {
    print STDERR "Usage: $ARGV[0] <orignal> <new>\n";
    exit 64;
}

open ORGINAL, "<$ARGV[0]";
open NEW, "<$ARGV[1]";

# Sanitize input, put into array
$tmp = join(" ",<ORGINAL>);
# Ignore case for the time beeing
$tmp = lc($tmp);
# Currently ignoring all punctations and newlines.
$tmp =~ tr/[\.:,;\n\r]//d;
@original = split(/[\t\ ]+/,$tmp);

# Sanitize input, put into array
$tmp = join(" ",<NEW>);
$tmp = lc($tmp);
$tmp =~ tr/[\.:,;\n\r]//d;
@new = split(/[\t\ ]+/, $tmp);

close ORGINAL;
close NEW;

# Bookkeeping values
$insert = 0;
$delete = 0;
$substitution = 0;

print "------------------\n";
# XXX: Some way of pretty print the actual text hits/misses
# Walktrough list comparing matches
$lasthit_orig = -1;
$lasthit_new = -1;
$newhit_orig = -1;
$newhit_new = -1;
foreach $np (0 .. $#new) {
    # Try to find word hit in next words
    $newhit_orig = -1;
    $end = $lasthit_orig + $LOOKFORWARD;
    $end = $#original if $end > $#original;
    foreach $op ($lasthit_orig .. $end) {
        if ( $new[$np] eq $original[$op] ) {
            $newhit_orig = $op;
            $newhit_new = $np;
            print "$new[$np] ";
            last;
        }
    }
    
    #Little hack to force processing the last words, on end of array, if none
    #found, deleting does not work properly, will be handled seperately
    if ($np == $#new and $newhit_orig == -1) {
        $newhit_orig = $#original + 1;
        $newhit_new = $#new + 1;
    }

    # We got a hit
    if ($newhit_orig >= 0) {
        # Calculate given diffences till (no including!) last hit
        $diff_orig = $newhit_orig - $lasthit_orig - 1;
        $diff_new = $newhit_new - $lasthit_new - 1;
        $diff_words = $diff_new - $diff_orig;

        # More means inserts, less deleting
        if ($diff_words > 0) {
           $insert += $diff_words;
        } else {
           $delete += abs($diff_words);
        }

        # The smallest number defines the substitutions
        if ($diff_new < $diff_orig) {
            $substitution += $diff_new;
        } else {
            $substitution += $diff_orig;
        }

        # Bookkeeping
        $lasthit_orig = $newhit_orig;
        $lasthit_new = $newhit_new;
    }
}

# Make sure to process last deletions
if ($#new < $#original) {
    $delete += $#original - $#new;
}

print "\n";
print "------------------\n";
print "Total orig : $#original\n";
print "Total new  : $#new\n";
print "------------------\n";
print "Insert     : $insert\n";
print "Delete     : $delete\n";
print "Subsitute  : $substitution\n";
print "------------------\n";
printf "WeR        : %.3f\n", ($insert + $delete + $substitution) / $#original, "\n";

exit 0;