Soundex Calculator – Turn Names Into Phonetic Codes
Soundex is a phonetic index: it compresses a name into one letter and three digits so that spellings which sound alike land on the same code. Robert and Rupert both become R163. That single idea, patented by Robert C. Russell and Margaret King Odell around 1918, is still doing real work a century later inside genealogy databases, library catalogues, record-linkage pipelines and the SOUNDEX() function of almost every SQL engine. This Soundex calculator encodes a name, a pair of names, or a whole pasted list — and, unlike a plain code dumper, shows you every step of the derivation.
How the Soundex algorithm works
The input is uppercased, accents are folded away and everything outside A–Z is stripped. The first letter is kept verbatim as the head of the code. Every remaining letter is then mapped to a digit: 1 = B F P V, 2 = C G J K Q S X Z, 3 = D T, 4 = L, 5 = M N, 6 = R. Vowels plus H, W and Y are not coded at all. Letters that share a digit and sit next to each other collapse into one, the sequence is cut after three digits, and anything shorter is padded with zeros — which is why Lee becomes L000.
The two rules everyone gets wrong
The first is H and W transparency. When two letters with the same digit are separated only by an H or a W, they still count as adjacent and produce a single digit. That is why Ashcraft encodes as A261 and not A226. A vowel between them does the opposite: it separates them so both are coded, giving Tymczak → T522. The second is first-letter suppression— if the second letter shares the first letter's digit it is dropped, so Pfister is P236 rather than P123.
Why your database disagrees
Implementations diverge, and the differences bite in production. SQL Server's SOUNDEX() does not apply the H/W rule the same way, so it reports A226 for Ashcraft. MySQL never truncates — SOUNDEX('Quadratically') returns the six-character Q36324. The textbook Simplified Soundex skips both special rules. And Refined Soundex abandons the four-character format entirely for a ten-group chart that produces far fewer collisions. The variant selector encodes your name under all of them at once so you can see exactly where the codes part company.
S530, while Carr and Karr never match because their initial letters differ. Use the code to shortlist candidate records, then confirm them with a second signal such as a date of birth or an edit-distance score.What you can do with the tool
Single mode encodes one name and prints a letter-by-letter trace table alongside a chip strip that shows which letters were kept, dropped, collapsed or truncated. Compare encodes two names, returns a match verdict and scores how close the codes are position by position. Bulk turns a pasted list into a Name → Soundex table with a collision rate, and Group clusters that list by code so duplicates surface immediately — the core move in any deduplication or record-linkage job. Reverse takes a code such as R163, decomposes it back into the candidate letter sets, and matches it against your own list or a built-in surname dictionary. Variants puts all five algorithms side by side.
Genealogy, deduplication and fuzzy search
The U.S. National Archives indexed the 1880–1920 federal censuses by Soundex precisely because enumerators wrote names by ear. If an ancestor has vanished from a census index, encoding the surname and searching by code recovers the mis-transcribed spellings. The same trick powers "sounds like" search over a customer or patient table: store the code in an indexed column, encode the query term, and match on equality. For surname particles such as Van, De, Mc or O', NARA sometimes filed a name both with and without the prefix — switch on prefix stripping to get both codes at once.
Limitations and alternatives
Soundex was tuned for English surnames and shows it: it is weak on Slavic, Chinese and Arabic transliterations, it is entirely dependent on the first letter being right, and it produces plenty of false positives. When precision matters more than recall, look at the Metaphone and NYSIIS codes shown alongside the Soundex result — both model English pronunciation more carefully. Every calculation here runs locally in your browser, so name lists never leave your device.