> metaphone | phonetic | sound <

Metaphone is a phonetic algorithm published by Lawrence Philips in 1990. It improves on Soundex by using more complex rules that better match English pronunciation patterns.

Metaphone uses more sophisticated rules and considers letter positions and combinations. It's more accurate for English words than Soundex, which was designed for surnames and uses simpler numeric codes.

Metaphone codes are commonly used in spell checkers, search engines, database deduplication, genealogy research, and any application that needs to match words that sound similar but may be spelled differently.

Yes, Metaphone is specifically designed for English pronunciation. For other languages, different phonetic algorithms like Cologne phonetic (German) or Caverphone (New Zealand English) may be more appropriate.

Lawrence Philips released three generations of his phonetic algorithm; pick based on your needs:

Metaphone (1990):
• 16 phonetic symbols: B X S K J T F H L M N P R 0 W Y.
• Single code output, e.g., Smith → SM0, Knight → NT.
• Open source, widely implemented (PHP metaphone(), Python jellyfish.metaphone).
• Handles English words much better than Soundex but still fails on non-English origins.

Double Metaphone (2000):
• Returns two codes per word: a primary and (sometimes) an alternate pronunciation. For Schmidt: primary XMT, alternate SMT — matches both the English "sh-mid" and German "s-mid" readings.
• Handles: English, German, Dutch, Slavic, Italian, Spanish, French, Chinese-transliterated surnames.
• Use when matching names from immigrant records, multilingual datasets, global customer data.
• Implementations: pip install doublemetaphone, npm install double-metaphone, Java Apache Commons Codec, PHP via Pear.

Metaphone 3 (2009):
• ~99% accuracy on common English words and names per Philips' claims.
• Commercial: requires a license from Philips (anthropomorphicsoftware.com). No free open source implementation.
• Use only if your product genuinely needs the edge over Double Metaphone.

Practical recommendation: start with Double Metaphone — it's free, open source, handles multilingual data, and for 95% of fuzzy-matching use cases it's better than Soundex, Metaphone 1, NYSIIS, or Caverphone. Upgrade to Metaphone 3 only if you have budget and measured accuracy gap.

Don't implement Metaphone from scratch — the rules (~30 cases of letter-combination handling) are easy to get subtly wrong. Use libraries:

Python — use jellyfish (fastest, C implementation) or fuzzy:
import jellyfish
jellyfish.metaphone('Smith') # 'SM0'
jellyfish.metaphone('Schmidt') # 'XMT'

# Double Metaphone
from doublemetaphone import doublemetaphone
doublemetaphone('Schmidt') # ('XMT', 'SMT')

JavaScript / Node.js:
const { doubleMetaphone } = require('double-metaphone');
doubleMetaphone('Schmidt'); // ['XMT', 'SMT']

// natural package for original Metaphone
const natural = require('natural');
natural.Metaphone.process('Smith'); // 'SM0'

Java: Apache Commons Codec
import org.apache.commons.codec.language.Metaphone;
import org.apache.commons.codec.language.DoubleMetaphone;
new Metaphone().encode("Schmidt"); // "SXMT"
DoubleMetaphone dm = new DoubleMetaphone();
dm.doubleMetaphone("Schmidt"); // "XMT"
dm.doubleMetaphone("Schmidt", true); // "SMT" (alternate)

PHP: built-in (original Metaphone only):
echo metaphone('Schmidt', 0); // 'XMT'
echo metaphone('Thompson', 0); // 'TMSN'

SQL — fuzzy search with Metaphone:
PostgreSQL + fuzzystrmatch extension:
CREATE EXTENSION fuzzystrmatch;
SELECT metaphone('Schmidt', 10);
SELECT dmetaphone('Schmidt'), dmetaphone_alt('Schmidt');

-- Find all customers whose name sounds like 'Smith'
SELECT * FROM customers
WHERE dmetaphone(last_name) = dmetaphone('Smith')
OR dmetaphone_alt(last_name) = dmetaphone('Smith');

Precompute and index the Metaphone column on large tables; otherwise the per-row function call prevents index usage.

Metaphone 和 Double Metaphone 都是为英语发音设计的，对汉字无效（算法只接受 A-Z 字母输入）。但如果你把中文名字 先拼音化，再用 Double Metaphone 做匹配，会比直接比较拼音字符串更宽容（例如能合并 "sh/s"、"zh/z" 的南方口音差异）。

各语言的语音算法选择：

Metaphone outputs a compact phonetic skeleton using 16 symbols. Reading common codes:

• SM0 — "S-M-TH" sound. Matches Smith, Smyth, Smythe (the 0 represents the th sound like in thin; not zero).
• XMT — "SH-M-T" sound. Matches Schmidt, Shmidt (German/Yiddish spelling).
• KF — "K-F" sound. Matches Cough, Coffee, Cofe (silent/combined letters normalized).
• NT — "N-T" sound. Matches Knight, Night, Nite (silent K dropped).
• JRJ — "J-R-J". Matches George, Jorge (Spanish spelling of the same name).
• TMSN — Thompson, Thomson, Tompson.
• FN — Phone, Fone.
• RF — Rough, Ruff, Rufe.

The 16 Metaphone symbols:
B X(sh) S K J T F H L M N P R 0(th) W Y

Key normalization rules:
• PH → F: phone becomes FN.
• CH → X (sh): church becomes XRX.
• C before E/I/Y → S: center becomes SNTR.
• C otherwise → K: cat becomes KT.
• G before E/I/Y → J: gem becomes JM.
• Silent letters dropped at word boundaries: K before N (Knight), G before N (Gnu, Gnome), P before N (Pneumonia), P before S (Psychology).
• Double consonants collapsed: ll → l, tt → t.
• Vowels dropped except as the first letter: Apple → APL, but Ball → BL.