You're confusing tr/// for s///; see perldoc perlop. But more to the point you probably want something like Text::CSV_XS to parse your CSV into an array, remove embedded commas as needed, and then use the same module to write it back out.

Tokenizer approach:

$input = '7374,726327,"76,237",32324,"21,342,857",23';
print("input:  $input\n");


$output = '';

foreach ($input) {
   if (/\G("[^"]+")/gc) {
      my $quoted = $1;
      $quoted =~ s/,//g;
      $output .= $quoted;
      redo;
   }

   # Unmatched quote.
   if (/\G"/gc) {
      $output .= '"';
      redo;
   }

   if (/\G([^"]+)/gc) {
      $output .= $1;
      redo;
   }
}

print("output: $output\n");
[download]

To avoid reinventing the wheel, I do recommed Text::CSV_XS or similar over my solution. Use it to seperated the fields, then remove the comma from every field with something like s/,//g foreach @fields;.

foreach sounds funny here.

It's using the fact that it aliases $_ so that you don't have to explicitly bind to $input each time. It's iterating over a list, it just happens to only have one element.

Of course some wonks might say that in addition to the spelled-for-pronounced-foreach and vice versa you've got this spelled-foreach-pronounced-with . . . :)

--
We're looking for people in ATL

s/("[^"]*")/ my $s= $1; $s =~ s-,--g; $s /ge;
[download]

use strict;
use warnings;

my $Orig = '7374,726327,"76,237",32324,"21,342,857",23';
my $Mod = $Orig;

1 while $Mod =~ s/((?:^|,)"[^"]*?),([^"]*?"(?:,|$))/$1$2/g;
print "$Orig => $Mod\n";

7374,726327,"76,237",32324,"21,342,857",23 => 7374,726327,"76237",3232
+4,"21342857",23
[download]

$string = '7374,726327,"76,237",32324,"21,342,857",23';
$string =~ s/("[^"]+")/my $tmp = $1; $tmp =~ s!,!!g; $tmp/eg;
print $string;
[download]

7374,726327,"76237",32324,"21342857",23
[download]

substr($str,index($str,'"'),rindex($str,'"')-length($str))=~y/,//d;
[download]