🚫 BadWords

High-performance profanity filter for Python, Rust, and JavaScript (WebAssembly)
with multilingual support and evasion detection.

---

---

Installation • Quick Start • Benchmarks • Supported Languages • Evasion Detection • Documentation

---

📖 Description

BadWords is a sophisticated profanity filtering library designed to clean up user-generated content. Unlike simple keyword matching, it uses similarity scoring, homoglyph detection, and transliteration to catch even the most cleverly disguised insults.

Architecture: The core is implemented in Rust for performance. Python provides a thin API layer with full type hints for IDE/linter support. The Rust library can also be used directly from Rust projects.

📦 Installation

Requirements

• Recommended: Python 3.13
• Minimum: Python 3.10+

Install via GitHub

pip install git+[https://github.com/FlacSy/badwords.git](https://github.com/FlacSy/badwords.git)

Install via PyPI

pip install badwords-py

---

⚡ Quick Start

Basic Initialization

from badwords import ProfanityFilter

# Initialize filter
p = ProfanityFilter()

# Load specific languages (e.g., English and Russian)
p.init(languages=["en", "ru"])

# Or load ALL 26+ supported languages
p.init()

Checking and Filtering Text

text = "Some very b4d text here"

# 1. Simple check (Returns Boolean)
is_bad = p.filter_text(text)
print(is_bad) # True

# 2. Censoring text (Returns String)
clean_text = p.filter_text(text, replace_character="*")
print(clean_text) # "Some very *** text here"

---

⏱ Benchmarks

CPU	GPU	RAM	OS
x86_64 i7 Intel® Core™ i7-10700KF × 16	NVIDIA GeForce RTX™ 3070	64 GB DDR4 3200MHz	Ubuntu 24.04.2 LTS

Rule-based matching (en+ru, match_threshold=1.0). Run: make bench

Scenario	Rust (badwords-core)	Python (badwords-py)
Clean text (no match)	~7.6 µs (~130 K/s)	~7.7 µs (~130 K/s)
Bad word (match)	~3.1 µs (~320 K/s)	~2.7 µs (~370 K/s)
Censor (replace)	~2.8 µs (~360 K/s)	~2.5 µs (~400 K/s)
5 texts batch	~15 µs (~330 K/s)	~16 µs (~310 K/s)

Python uses Rust via PyO3, overhead minimal.

vs glin-profanity

Rule-based mode, en+ru. Run: make bench-compare (requires pip install glin-profanity)

Scenario	BadWords	glin-profanity
Clean text	~7 µs (~140 K/s)	~4.4 ms (~230/s)
Bad word	~1.3 µs (~770 K/s)	~0.2 ms (~5 K/s)
Censor	~1.8 µs (~560 K/s)	~1.4 ms (~700/s)
5 texts batch	~16 µs (~310 K/s)	~10 ms (~500/s)

BadWords is ~100–600× faster (Rust core vs pure Python).

ML mode

pip install glin-profanity[ml] + make bench-compare. 100 iter each.

Scenario	BadWords ML (ONNX)	glin transformer
Clean text (43 chars)	~6.5 ms (~150/s)	~27 ms (~37/s)
Bad word (8 chars)	~4.6 ms (~220/s)	~21 ms (~47/s)
5 texts batch (82 chars)	~24 ms (~210/s)	~107 ms (~47/s)

BadWords ML (XLM-RoBERTa) ~3–4× faster than glin transformer.

---

🛠 Methods & API

filter_text(text, match_threshold=1.0, replace_character=None)

The core method of the library.

Parameter	Type	Default	Description
text	str	Required	Input text to check.
match_threshold	float	1.0	Similarity threshold (1.0 = exact match, 0.95 = fuzzy).
replace_character	str/None	None	If provided, returns censored string. If None, returns bool.

Warning

Performance Tip: Using match_threshold < 1.0 enables fuzzy matching which is slower. Use 1.0 for high-traffic real-time filtering, or 0.95 for a good balance.

---

🧩 Advanced Evasion Detection

Standard filters are easy to bypass. BadWords is built to detect:

• Homoglyphs: Detects hеllo (using Cyrillic 'е') or h4llo (numbers).
• Transliteration: Automatically handles mapping between Cyrillic and Latin alphabets.
• Normalization: Strips diacritics, special characters, and decorative Unicode symbols.
• Similarity Analysis: Uses fuzzy matching to find words with deliberate typos.

Examples of detected evasions:

_filter.filter_text("hеllо")  # Mixed alphabets (Cyrillic + Latin) -> DETECTED
_filter.filter_text("h3ll0")  # Character substitution -> DETECTED
_filter.filter_text("h⍺llo")  # Mathematical/Greek symbols -> DETECTED
_filter.filter_text("привет") # Transliterated matches -> DETECTED

---

🌍 Supported Languages

BadWords supports 25 languages out of the box:

Code	Language	Code	Language	Code	Language
en	English	ru	Russian	ua	Ukrainian
de	German	fr	French	it	Italian
sp	Spanish	pl	Polish	cz	Czech
ja	Japanese	ko	Korean	th	Thai
br	Portuguese (BR)	da	Danish	du	Dutch
fi	Finnish	gr	Greek	hu	Hungarian
in	Indonesian	lt	Lithuanian	no	Norwegian
po	Portuguese	ro	Romanian	sw	Swedish
tu	Turkish

Use p.get_all_languages() in code. Full list with word counts: badwords.flacsy.dev

---

🚀 Full Integration Example

from badwords import ProfanityFilter

def monitor_chat():
    # Setup for a global chat
    profanity_filter = ProfanityFilter()
    profanity_filter.init(["en", "ru", "de"])
    
    # Custom project-specific banned words
    profanity_filter.add_words(["spam_link_v1", "scam_bot_99"])

    user_input = "Hey! Check out this b.a.d.w.o.r.d"
    
    # Moderate with high accuracy
    is_offensive = profanity_filter.filter_text(user_input, match_threshold=0.95)
    
    if is_offensive:
        print("Message blocked: Contains restricted language.")
    else:
        # Proceed with processing
        pass

if __name__ == "__main__":
    monitor_chat()

---

🦀 Rust API (badwords-core)

Published on crates.io:

[dependencies]
badwords-core = "2"

use badwords_core::{ProfanityFilter, default_resource_dir};

let resource_dir = default_resource_dir();
let mut filter = ProfanityFilter::new(&resource_dir, true, true, true, true);
filter.init(None).unwrap();
filter.add_words(&["custom".to_string()]);
let (found, _) = filter.filter_text("hello", 1.0, None);

🌐 WebAssembly (JavaScript/TypeScript)

Same Rust code for browser and Node.js, compiled to WASM.

Build

# Browser
make wasm

# Node.js
make wasm-nodejs

Frontend (browser)

<script type="module">
  import init, { ProfanityFilter } from './path/to/badwords_wasm.js';
  await init();
  const filter = new ProfanityFilter();
  console.log(filter.isBad('text'));      // boolean
  console.log(filter.censor('text', '*')); // string
</script>

Backend (Node.js)

const { ProfanityFilter } = require('badwords-wasm');
const filter = new ProfanityFilter();
filter.isBad('hello');           // false
filter.censor('bad word', '*');  // "*** word"
filter.addWords(['custom']);

Optional languages (npm)

Built-in: en and ru. Additional languages via @badwords/languages:

npm install badwords-wasm @badwords/languages

import init, { ProfanityFilter } from 'badwords-wasm';
import de from '@badwords/languages/de';
import ua from '@badwords/languages/ua';

await init();
const filter = new ProfanityFilter();
filter.addWords(de);
filter.addWords(ua);

Available: br, cz, da, de, du, en, fi, fr, gr, hu, in, it, ja, ko, lt, no, pl, po, ro, ru, sp, sw, th, tu, ua. See @badwords/languages.

Examples: examples/wasm/browser/, examples/wasm/node/

🔧 Building from source

Requires: Rust, Python, maturin

python -m venv .venv && source .venv/bin/activate  # Linux/macOS
pip install maturin
make develop
# or: cd python && maturin build && pip install target/wheels/badwords_py-*.whl

🌐 WebAssembly (browser & Node.js)

Build the WASM package (requires wasm-pack):

cargo install wasm-pack
make wasm

Output: rust/badwords-wasm/pkg/ (npm package badwords-wasm)

• Browser: Use the generated JS with a bundler or static server. See examples/wasm/browser/
• Node.js: import init, { ProfanityFilter } from 'badwords-wasm' after npm install. See examples/wasm/node/
• Publish to npm: make wasm or make wasm-nodejs, then make npm-publish
• Optional languages: @badwords/languages — make lang-packages then make npm-publish-languages

📚 Documentation

Full documentation (Python, Rust, JavaScript) with examples and API reference: badwords.flacsy.dev (EN / RU).

🤝 Contributing

Contributions are what make the open-source community an amazing place to learn, inspire, and create.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

📄 License

Distributed under the MIT License. See LICENSE for more information.

_{Developed with ❤️ by FlacSy}