thai_profanity_filter

A Thai-aware profanity filter for Dart — detect, censor, and score Thai bad words. Pure Dart, dependency-free, works everywhere (mobile, web, desktop, server, CLI).

Thai has no spaces between words, so a naive "does the string contain a bad word" check both misses evasions (ค ว ย, เหี้ยยยย) and flags innocent words (หีบ "chest" contains หี). This package handles both: it normalizes away common evasions and uses an allow-list so clean words that merely contain a bad substring are not flagged.

Features

• Detect / score / censor — hasProfanity, isClean, severityOf, findMatches (with positions in the original string), and censor.
• Severity tiers — mild (rude pronouns like กู/มึง), moderate (clearly vulgar), severe (sexual / strongly offensive). Filter by a minimum severity.
• Evasion handling — collapses repeated characters (เหี้ยยยย → เหี้ย), joins spaced-out letters (ค ว ย → ควย), and strips zero-width characters.
• False-positive guard — a bundled allow-list of clean words that contain a bad substring (e.g. หีบ, ตระกูล, แกงกะหรี่) so they are never flagged.
• Fully customizable — add your own words, unblock defaults, extend the allow-list, all per filter instance.

Install

dependencies:
  thai_profanity_filter: ^0.1.0

Quick start

import 'package:thai_profanity_filter/thai_profanity_filter.dart';

void main() {
  final filter = ThaiProfanityFilter();

  filter.hasProfanity('ไอ้ควย');        // true
  filter.isClean('สวัสดีครับ');          // true
  filter.hasProfanity('เปิดหีบสมบัติ');   // false  (หีบ is allow-listed)

  // Evasion is handled by default:
  filter.hasProfanity('เหี้ยยยย');       // true
  filter.hasProfanity('ค ว ย');         // true

  // Censor, preserving the surrounding text:
  filter.censor('เอ๊ย ไอ้ควย');          // 'เอ๊ย ไอ้***'

  // Inspect what matched, and where (indices into the original string):
  for (final m in filter.findMatches('มึงมันเหี้ย')) {
    print('${m.word} [${m.start}..${m.end}] ${m.severity}');
  }
}

Severity & the default threshold

Severity is ordered mild < moderate < severe. By default the filter flags everything from mild up, so the rude pronouns กู / มึง are caught. If you only care about clearly-vulgar language and want the fewest false positives, raise the threshold:

final strict = ThaiProfanityFilter(minSeverity: Severity.moderate);
strict.isClean('มึงกับกู'); // true  — mild pronouns ignored
strict.hasProfanity('ควย');  // true  — still caught

Customizing

final filter = ThaiProfanityFilter(
  extraWords: {'บักหำ': Severity.mild},  // add your own
  removeWords: {'กู'},                    // unblock a default
  allowList: {'ระยำกิจ'},                  // never flag this clean word
  minSeverity: Severity.moderate,
  detectEvasion: true,
);

Notes & limitations

• It's a heuristic, not a moderator. Thai profanity is deeply context-dependent (เหี้ย is also an animal, ควาย/สัตว์ are animals). The default list deliberately leaves out the most ambiguous dual-use words and relies on an allow-list for the rest; you will still want human review for anything high-stakes.
• The allow-list is not exhaustive. A rare clean word that contains a bad substring and isn't listed may be flagged — add it via allowList.
• Evasion vs. false positives. detectEvasion joins spaced letters, which in rare cases can join two clean words into a flagged sequence. Turn it off for exact-only matching.
• The bundled word list is a starter set; extend it for your product.

License

MIT © 2026 MaIII Themd (ultramcu)