Remove accents and diacritics

Removing accents sounds like one job and is really two, which is why so many tools do half of it.

The first half is Unicode normalisation. Most accented Latin letters are stored as a base letter followed by one or more combining marks, or can be decomposed into that form on demand. Normalising to NFKD performs that split, and deleting the combining marks then leaves the plain base letter behind. This is what turns Crème Brûlée into Creme Brulee, and it handles the great majority of French, Spanish, Portuguese, Italian, Czech, Turkish and Vietnamese text correctly.

The second half is the letters that have no decomposition at all, because they are not a base plus a mark. They are letters in their own right that happen to look like a modified Latin letter to an English-speaking reader. The German eszett, the Scandinavian o-slash and ash, the Polish l-stroke, the Croatian d-stroke, and the Icelandic thorn and eth all fall into this group. Normalisation does nothing to them, so a tool built on normalisation alone hands back Straße, Ørsted and Łódź unchanged while reporting success.

FileGizmo names those letters explicitly and converts each one to the substitution its own language uses when the letter is unavailable. Two of them expand rather than swap: the eszett becomes ss and ash becomes ae, exactly as a German or Danish keyboard fallback would write them. Straße becomes Strasse, Ørsted becomes Orsted, and Łódź becomes Lodz.

This matters most where text has to match something else. A customer name imported into a system that stores ASCII, an email address derived from a person’s name, a filename that has to survive an old file server, a URL slug, a sort order that has to agree with a database collation. In every one of those cases a single surviving non-ASCII letter is the difference between a match and a silent failure that surfaces weeks later as a duplicate record.

What the tool deliberately does not do is transliterate other writing systems. Converting Greek, Cyrillic or Chinese to Latin letters is a translation decision with several defensible answers and no correct one, and picking silently would damage text the person did not intend to change. Those characters pass through untouched.

The result reports how many characters were replaced, so a run that changes nothing says so rather than looking identical to a run that worked. Everything happens in this browser tab, and nothing is uploaded.