This approach gets a bit complicated when one works with texts in languages that use many letters not present in the 32..127 range, though.