kham-wasm

WebAssembly bindings for the kham Thai NLP engine, built with wasm-bindgen.

Install

npm install kham-wasm

Quick start

import init, { segment, segment_fts } from "kham-wasm";

await init();

// Segmentation
const words = segment("กินข้าวกับปลา");
// ["กิน", "ข้าว", "กับ", "ปลา"]

// FTS pipeline — POS, NE, stopwords, romanization
for (const t of segment_fts("นายกรัฐมนตรีกินข้าว")) {
    console.log(t.text, t.pos, t.ne, t.is_stop, t.roman);
}

// Number conversion
number_to_thai_word(1234n);    // "หนึ่งพันสองร้อยสามสิบสี่"
thai_word_to_number("สองล้าน"); // "2000000" (empty string = not a number)

API reference

Segmentation

segment(text: string) → string[]

Segment Thai text and return an array of token strings. Whitespace excluded.

segment("กินข้าวกับปลา");
// ["กิน", "ข้าว", "กับ", "ปลา"]

segment_tokens(text: string) → Token[]

Segment and return Token objects with span information.

for (const t of segment_tokens("ธนาคาร100แห่ง")) {
    console.log(t.text, t.char_start, t.char_end, t.kind, t.confidence);
}
// ธนาคาร  0   6  Thai    0.95
// 100     6   9  Number  1
// แห่ง    9  13  Thai    1

segment_above_confidence(text: string, min_confidence: number) → Token[]

Returns only tokens whose confidence meets the threshold. Uses the streaming iterator internally.

const toks = segment_above_confidence("ธนาคาร100แห่ง", 0.9);
for (const t of toks) {
    console.log(t.text, t.confidence);
}

segment_fts(text: string) → FtsToken[]

Full NLP pipeline: normalize → segment → NE → stopwords → POS → synonyms → romanization. Returns FtsToken objects.

for (const t of segment_fts("นายกรัฐมนตรีกินข้าว")) {
    console.log(`${t.text.padEnd(10)} pos=${t.pos} ne=${t.ne} stop=${t.is_stop}`);
}

Romanization

romanize(text: string) → RomanToken[]

Segment and return each token paired with its RTGS romanization.

for (const t of romanize("กินข้าว")) {
    console.log(t.text, "→", t.roman);
}
// กิน  → kin
// ข้าว → khao

romanize_sentence(text: string) → string

Segment and romanize the entire text in one call. Thai and Named tokens are converted to RTGS; numbers, Latin, punctuation, and whitespace pass through unchanged.

romanize_sentence("กินข้าวกับปลา");      // "kin khao kap pla"
romanize_sentence("กินข้าว 100 บาท");    // "kin khao 100 baht"
romanize_sentence("hello โลก");          // "hello lok"

Normalization

normalize(text: string) → string

Apply two-rule Thai normalization:

1. Duplicate tone marks — consecutive tone marks collapsed to the last one.
2. Sara Am composition — nikhahit (อํ U+0E4D) + sara aa (อา) → sara am (อำ U+0E33).

normalize("ข้้าว");  // "ข้าว"  (deduplicate tone)
normalize("กํา");   // "กำ"    (compose sara am)

Sentence splitting

split_sentences(text: string) → Sentence[]

Split text into sentence spans. Boundaries: Thai markers (ฯ ๚ ๛), newlines, !, ?, . followed by a space.

for (const s of split_sentences("กินข้าวแล้ว! ดื่มน้ำด้วย")) {
    console.log(s.text, s.char_start, s.char_end);
}

Spell checking

spell_suggestions(word: string, max_n?: number) → SpellSuggestion[]

Return up to max_n spelling suggestions for word, ranked by edit distance then phonetic similarity then corpus frequency.

for (const s of spell_suggestions("กีนข้าว")) {
    console.log(s.word, s.edit_distance, s.soundex_match, s.freq_score);
}
// กินข้าว  1  true  1342

spell_did_you_mean(word: string) → string

Return the single best correction for word, or "" (empty string) if the word is already in the dictionary.

spell_did_you_mean("กีนข้าว");  // "กินข้าว"
spell_did_you_mean("กินข้าว");  // ""  (already correct)
spell_did_you_mean("");          // ""

spell_correct_text(text: string) → string

Segment text and replace every Unknown token (≥ 2 characters) with its best spelling correction. Known tokens pass through unchanged.

spell_correct_text("ผมกีนข้าวกับปลา");  // "ผมกินข้าวกับปลา"
spell_correct_text("กินข้าว");           // "กินข้าว"  (no change)

Keyword extraction

extract_keywords(text: string, max_n?: number) → Keyword[]

Extract the top max_n keywords from text by TF × IDF-proxy score. Stopwords and single-character tokens are excluded.

for (const kw of extract_keywords("นายกรัฐมนตรีประกาศนโยบายเศรษฐกิจ", 5)) {
    console.log(kw.word, kw.score, kw.count);
}

extract_phrases(text: string, max_n?: number) → Keyword[]

Extract the top max_n bigram and trigram keyphrases from adjacent content tokens, scored by TF × average-IDF. Returns the same Keyword type as extract_keywords.

const text = "การพัฒนาซอฟต์แวร์เป็นสิ่งสำคัญในยุคดิจิทัล";
for (const p of extract_phrases(text, 5)) {
    console.log(p.word, p.score, p.count);
}
// การพัฒนาซอฟต์แวร์  0.842  1

Soundex (phonetic encoding)

soundex_word(word: string, algo?: string) → string

Encode a Thai word using a phonetic algorithm.

| algo | Groups | Length | Notes | |---|---|---|---| | "lk82" (default) | 12 | 4 chars | Royal Institute 1982, most common | | "udom83" | 14 | 4 chars | Finer sibilant distinctions | | "metasound" | — | 3 chars/syllable | Per-syllable encoding |

soundex_word("กาน");              // "1600"
soundex_word("กาน", "udom83");    // "1900"
soundex_word("กาน", "metasound"); // "112"

sounds_like(a: string, b: string, algo?: string) → boolean

sounds_like("กาน", "ขาน");             // true  (same lk82 group)
sounds_like("ลาน", "ราน", "udom83");   // false (ล/ร split in udom83)

thai_english_soundex(word: string) → string

Thai–English cross-language soundex (Suwanvisat & Prasitjutrakul 1998). Accepts both Thai script and ASCII input.

thai_english_soundex("Robert");  // "671763"
thai_english_soundex("โรเบิร์ต");  // same prefix as "Robert"

sounds_like_cross_lang(a: string, b: string) → boolean

sounds_like_cross_lang("สมชาย", "Somchai");  // true
sounds_like_cross_lang("Robert", "Rupert");   // true

Number conversion

thai_digits_to_ascii(text: string) → string

Convert Thai digit characters (๐–๙) to ASCII. Other characters unchanged.

thai_digits_to_ascii("ราคา ๑๒๓ บาท");    // "ราคา 123 บาท"
thai_digits_to_ascii("ธนาคาร๑๐๐แห่ง");  // "ธนาคาร100แห่ง"

number_to_thai_word(n: bigint) → string

Convert a non-negative integer to its Thai cardinal word representation.

number_to_thai_word(0n);          // "ศูนย์"
number_to_thai_word(21n);         // "ยี่สิบเอ็ด"
number_to_thai_word(1_000_000n);  // "หนึ่งล้าน"

thai_word_to_number(text: string) → string

Parse a Thai cardinal number word. Returns "" (empty string) for non-number input, or the numeric value as a decimal string for valid input.

thai_word_to_number("หนึ่งร้อยยี่สิบสาม");  // "123"
thai_word_to_number("สองล้าน");              // "2000000"
thai_word_to_number("กินข้าว");              // ""

number_to_baht_text(baht: bigint, satang: number) → string

Render a Baht amount as Thai currency text (satang must be 0–99).

number_to_baht_text(100n, 0);   // "หนึ่งร้อยบาทถ้วน"
number_to_baht_text(21n, 50);   // "ยี่สิบเอ็ดบาทห้าสิบสตางค์"

parse_baht_text(text: string) → BahtResult

Parse a Thai Baht currency string. Check result.valid before using.

const r = parse_baht_text("หนึ่งร้อยบาทถ้วน");
if (r.valid) {
    console.log(r.baht, r.satang);  // 100n  0
}

Classes

Token

| Field | Type | Description | |---|---|---| | text | string | Token text | | byte_start / byte_end | number | UTF-8 byte offsets | | char_start / char_end | number | Unicode scalar-value offsets (use for string slicing) | | kind | string | "Thai" | "Latin" | "Number" | "Punctuation" | "Emoji" | "Whitespace" | "Unknown" | "Person" | "Place" | "Org" | | confidence | number | 0 (Unknown token) … 1 (unambiguous dict match) |

FtsToken

| Field | Type | Description | |---|---|---| | text | string | Token text (normalized) | | position | number | Ordinal position in non-whitespace sequence (0-based) | | kind | string | Same values as Token.kind | | is_stop | boolean | true if in the built-in stopword list | | roman | string | RTGS romanization (equals text for non-Thai / OOV) | | pos | string \| null | POS tag: "Noun" | "Verb" | "Adj" | "Adv" | "Particle" | "ProperNoun" | "Pronoun" | "Numeral" | "Classifier" | "Conjunction" | "Auxiliary" | "Determiner" | "Preposition" | | ne | string \| null | NE category: "Person" | "Place" | "Org" | | synonyms | string[] | Synonym / number-normalization expansions | | trigrams | string[] | Character trigrams (Unknown tokens only) |

RomanToken

| Field | Type | Description | |---|---|---| | text | string | Original token text | | roman | string | RTGS romanization |

Sentence

| Field | Type | Description | |---|---|---| | text | string | Sentence text | | char_start / char_end | number | Unicode scalar-value offsets |

BahtResult

| Field | Type | Description | |---|---|---| | valid | boolean | true if the input was a valid Baht string | | baht | bigint | Whole baht amount (only meaningful when valid is true) | | satang | number | Satang 0–99 (only meaningful when valid is true) |

SpellSuggestion

| Field | Type | Description | |---|---|---| | word | string | Suggested word | | edit_distance | number | Levenshtein distance from the input (1 or 2) | | soundex_match | boolean | true if the lk82 soundex code matches the input | | freq_score | number | TNC corpus frequency (higher = more common) |

Keyword

| Field | Type | Description | |---|---|---| | word | string | Keyword text | | score | number | TF × IDF-proxy score | | count | number | Occurrence count in the input text |

Build from source

git clone https://github.com/preedep/kham
cd kham
wasm-pack build kham-wasm --target web
# → kham-wasm/pkg/

For Node.js:

wasm-pack build kham-wasm --target nodejs

Links

• kham.io — live demo & full documentation
• GitHub
• kham-core on crates.io
• API reference