What is a Soundex code?

It is a 4-character code (one letter plus three digits) that represents how a name sounds rather than how it is spelled. Names that sound alike usually get the same code, which makes it useful for fuzzy matching.

Which Soundex variant does this use?

It implements American Soundex as defined by the US Census Bureau and the National Archives (NARA), including the H/W transparency rule and vowel separator handling.

Why do Robert and Rupert give the same code?

Both encode to R163. Soundex keeps the first letter and codes B/P/F/V as 1 and R/T as 6, so the differing vowels and the B-vs-P distinction are deliberately ignored.

Is Soundex case sensitive?

No. The encoder uppercases the input and strips non-letters first, so "O'Brien" and "obrien" produce the same result.

What are the limits of Soundex?

It was designed for English surnames and is fairly coarse. Metaphone or NYSIIS are usually more accurate, and none of them handle non-English names well.

Soundex Encoder — Gera Tools

What this tool does

Soundex is a phonetic algorithm that turns a name into a short code so that words which sound similar map to the same value. It was created for the US Census to group surnames despite spelling variations and misspellings. This encoder produces the standard 4-character Soundex code — one retained first letter plus three digits — entirely in your browser.

How it works

The algorithm keeps the first letter of the name, then codes the remaining letters by sound:

1 = B F P V
2 = C G J K Q S X Z
3 = D T
4 = L
5 = M N
6 = R
ignored = A E I O U Y H W

Three rules refine the result. First, two adjacent letters that share a code count only once (and a letter immediately following the first letter is dropped if it shares the first letter’s code). Second, the letters H and W are transparent: two coded letters separated by H or W still collapse as if adjacent. Third, vowels (A, E, I, O, U, Y) act as separators, so coded letters split by a vowel are kept distinct. The output is the first letter plus three digits, zero-padded if short and truncated if long — for example Robert and Rupert both give R163.

Example and notes

Try Ashcraft, which correctly yields A261 under the NARA rules (the silent-collapse behaviour around C/R), and Tymczak, which gives T522. A matching Soundex code confirms two names are phonetically similar; it does not prove they refer to the same person. Use it as a candidate-generation step for record linkage or search, then verify matches with stricter comparison. Everything is computed locally — no names are transmitted.