@oomfware/lang-detect

natural language detection library.

npm install @oomfware/lang-detect

usage

call initialize() once to load model weights via fetch(), then use detect() synchronously on any text. results are sorted by probability, highest first.

import { initialize, detect } from '@oomfware/lang-detect';

await initialize();

const results = detect('the quick brown fox jumps over the lazy dog');
console.log(results[0]); // -> ['eng', 0.98]

mixed-script text returns detections for each script present, with probabilities scaled by proportion:

const results = detect('Hello Мир');
// -> [['eng', ...], ['rus', ...]]

variants

two variants are available, trading accuracy for smaller weights:

subpath	weights	accuracy
(default)	74.9 kB	97.4%
/lite	48.3 kB	97.2%

import { initialize, detect } from '@oomfware/lang-detect/lite';

supported languages

50 languages across Latin, Cyrillic, Arabic, Devanagari, CJK, and unique-script families.

code	language	code	language	code	language
afr	Afrikaans	hau	Hausa	por	Portuguese
ara	Arabic	heb	Hebrew	ron	Romanian
aze	Azerbaijani	hin	Hindi	run	Rundi
bel	Belarusian	hrv	Croatian	rus	Russian
ben	Bengali	hun	Hungarian	slk	Slovak
bul	Bulgarian	hye	Armenian	spa	Spanish
cat	Catalan	ind	Indonesian	srp	Serbian
ces	Czech	isl	Icelandic	swe	Swedish
ckb	Central Kurdish	ita	Italian	tgl	Tagalog
cmn	Mandarin Chinese	jpn	Japanese	tur	Turkish
dan	Danish	kat	Georgian	ukr	Ukrainian
deu	German	kaz	Kazakh	vie	Vietnamese
ell	Greek	kor	Korean
eng	English	lit	Lithuanian
est	Estonian	mar	Marathi
eus	Basque	mkd	Macedonian
fin	Finnish	nld	Dutch
fra	French	nob	Norwegian Bokmål
pes	Persian	pol	Polish