goldenphonetic

Fast, byte-identical-to-jellyfish phonetic encoders. A thin PyO3 wheel over the pyo3-free goldenphonetic-core Rust crate, with zero runtime dependencies.

import goldenphonetic as gp

gp.soundex("Robert")               # -> 'R163'   (American Soundex)
gp.metaphone("Thompson")           # -> '0MPSN'  (original Metaphone, Philips 1990)
gp.nysiis("Catherine")             # -> 'CATARAN'
gp.match_rating_codex("Byrne")     # -> 'BYRN'   (Match Rating Approach codex)
gp.match_rating_comparison("Byrne", "Boern")   # -> True

Every function is a byte-for-byte drop-in for the corresponding function in the Python jellyfish package (>= 1.0), including jellyfish's exact uppercasing and Unicode (NFKD / grapheme) handling — verified against jellyfish over a 6,000-input + 2,500-pair fuzz corpus (goldenphonetic-core/scripts/check_phonetic_parity.py), with no jellyfish at runtime.

Functions

Error semantics (mirrors jellyfish exactly)

• match_rating_codex raises ValueError when the input contains any non-alphabetic, non-space character ("O'Brien", "Smith-Jones", "123").
• match_rating_comparison never raises — it returns None when the two codices can't be compared: a codex-length difference of 3 or more (match_rating_comparison("Tim", "Timothy") is None), or either input rejected by the codex (non-alphabetic).

Unicode

Inputs are uppercased with the same algorithm jellyfish uses, then soundex/metaphone apply full NFKD decomposition (vendored as generated static data — no unicode-normalization dependency) and nysiis segments grapheme clusters (UAX-29 GB9 over a vendored combining-mark table). Accented names match in both precomposed and decomposed form.

Part of the golden suite.