Sorry, but you can't consistently parse data that is THIS inconsistent. Blank lines are easy to ignore, but how can you ignore lines with "noise" at the start without a solid way to denote the start of your data?

Try your best to clean up the incoming data. Until then, here are some parsing tricks that might help you keep some of your code maintainable (note I didn't say fast).

my %KEY_PARSER = (
                  "number" => {
                      START_COMMAND => qr{number:\s*}i,
                      VALUE_MATCH => qr{\d+},
                  },
                  "hair color" => {
                      START_COMMAND => qr{hair colou?r:\s*}i,
                      VALUE_MATCH => qr{[\w\s]+},
                  },
                  "height" => {
                      START_COMMAND => qr{height:\s*}i,
                      VALUE_MATCH => qr{\d+},
                  },
                  "weight" => {
                      START_COMMAND => qr{weight:\s*}i,
                      VALUE_MATCH => qr{\d+},
                  },
                  );

foreach my $line (grep {/\w/} <DATA>) {
    foreach (keys %KEY_PARSER) {
        while ($line =~ /$KEY_PARSER{$_}{START_COMMAND}/) {
            $line =~ s/($KEY_PARSER{$_}{START_COMMAND})\s*($KEY_PARSER
+{$_}{VALUE_MATCH})//;
            next unless $2;
            my ($key,$value) = ($1,$2);
            chomp ($key,$value);
            print "Found KEY: $key = $value\n";
        }
    }
}
[download]

Crap, even when munged by the magical Perl, still smells like crap.